JP5104496B2 - Diffusion prediction system, method and program, and impact estimation system, method and program - Google Patents

Description

  The present invention relates to a spread prediction system, a spread prediction method, and a spread prediction program that make predictions regarding the spread of a product or service, and a customer of the product or service in a certain period causes other people to use the product or service. The present invention relates to an influence degree estimation system, an influence degree estimation method, and an influence degree estimation program for estimating a degree.

  When providing goods or services to customers, it is necessary to predict the transition of diffusion in order to estimate the cost-effectiveness of goods or services and the timing and amount of resources to be procured. Here, the product includes not only a finished product such as a product but also a minimum unit for maintaining the function of the product. The transition of the spread of products and services can be shown by the spread rate (or the number of sales) with respect to time.

  An example of a prediction device applicable to such a spread prediction is described in Patent Document 1. The prediction device described in Patent Literature 1 includes an input unit, an event content storage unit, a cumulative number counting unit, a prediction condition storage unit, a regression analysis unit, a distribution estimation unit, and an output unit. . The input section includes the past history such as the date and number of occurrences of the event to be predicted for the phenomenon of interest, and the range and prediction of values such as the upper and lower limits of the event to be predicted in the event to be predicted This is an input means for inputting a prediction condition represented by a point in time. The regression analysis unit uses a logistic curve model (or another curve model) to estimate the parameters describing the phenomenon to be noted so as to apply to the transition of the actual results, and based on the estimated parameters, Obtain a prediction curve that is a transition.

  Patent Document 2 describes a time-series data classification / prediction device that generates decision tree learning data. The time-series data classification / prediction device described in Patent Document 2 generates decision tree learning data including information on a plurality of times.

  Further, Patent Document 3 describes an analysis apparatus using customer attribute data as master data.

JP 2004-78780 A JP-A-6-96052 JP 2001-229150 A

  The time when a customer starts using a product or service depends on the customer's characteristics. For example, customers can be categorized into categories such as innovators, early adapters, early majority.

  As described above, the time when a customer starts using a product or service varies depending on the characteristics of the customer, so even when trying to predict the start of use of a product or service for each individual customer, a logistic model such as a logistic curve is used. In the prediction method that performs parameter fitting, the characteristics of individual customers are not reflected, and it is impossible to predict the use start time of products and services for each customer.

  In addition, it is preferable for a provider of a product or service to know the degree to which a person who already uses the product or service urges others to use the product or service. For example, knowing how much the so-called word-of-mouth has been activated by advertising activities and sales promotion activities, and the degree of motivation for non-users to start using the products and services. It is preferable that Here, the degree to which a customer of a product or service urges others to use the product or service within a certain period is called an influence level. The degree of influence can also be said to be the degree of acceptability for accepting a product due to the influence from a person who has not started using the product or service in a certain period.

  Therefore, an object of the present invention is to provide a spread prediction system, a spread prediction method, and a spread prediction program capable of predicting for each customer when a customer starts using a product or service. It is another object of the present invention to provide an influence degree estimation system, an influence degree estimation method, and an influence degree estimation program that can estimate the influence degree of a product or service according to time.

  The spread prediction system of the present invention indicates that the customer or service starts when the product or service is started, and indicates that the product or service is not used when the product or service is not used. Using a customer database that stores customer data that is a set of individual data for each customer including use start information and one or more items representing customer attributes other than the use start information, and determination using the customer database A test data generation unit that generates test data that includes individual data of customers who have not started using the product or service at the current time that defines the target time; and the use of the product or service is started at the current time Label each customer's individual data with a first label, and second customer individual data that has not started using the goods or services. The number of individual data labeled with the first label and the first label varies depending on the degree of influence, which is the degree to which the customer of the product or service will encourage others to use the product or service within a certain period of time Classification that is a rule for determining which of the first label and the second label to label the individual data of the customer from the item representing the attribute of the customer, and the learning data generation unit that generates the learning data that is the processed data A test data label for determining a label for each individual data in the test data from a classifier generating unit that generates a classifier based on the learning data, and each individual data item in the classifier and the test data And a determination unit.

  The influence estimation system of the present invention represents the use start time of the product or service when the customer starts using the product or service, and is not used when the product or service is not used. A customer database that stores customer data, which is a set of individual data for each customer, including usage start information representing the effect and one or more items representing customer attributes other than the usage start information; A current time used as an estimation target time of the degree of influence, which is a degree that a customer of the product or service urges others to use the product or service, and a time interval for designating a predetermined time before the current time A plurality of provisional influence degrees that are candidates for influence degree are input, a previous time that is a time before the time interval from the current time is calculated, and the customer database is used to The first label is labeled on the individual data of the customer who has started using the product or service at the current time, and the second label is labeled on the individual data of the customer who has not started using the product or service. A current time data generating unit that generates time data, and for each temporary influence degree, the first label is labeled on individual data of a customer who has started using the product or service at the previous time, and the product or service Before generating the previous time data, which is data obtained by labeling the individual data of customers who have not started using the second label with the second label, and changing the number of individual data labeled with the first label according to the temporary influence degree For each time data group generation unit and each previous time data, the customer's individual data is labeled with either the first label or the second label from the item representing the customer's attribute. A time classifier group generating unit that generates a classifier that is a rule for determining the time based on the previous time, and for each individual classifier generated for each previous time data, the classifier and the current time data For each individual data item, an error group calculation unit that predicts a label to be labeled on the individual data in the current time data, calculates an error between the prediction result and the current time data, and for each classifier Among the calculated errors, the minimum error is identified, and when there is one temporary influence level corresponding to the minimum error, the temporary influence level is determined as the influence level, and the temporary influence corresponding to the minimum error is determined. When there are a plurality of degrees, an influence degree calculation unit that determines the influence degree based on the plurality of provisional influence degrees is provided.

  Further, the spread prediction system of the present invention indicates that the customer or service starts when the customer starts using the product or service, and indicates that the product or service is unused when the product or service is not used. Using a customer database that stores customer data that is a set of individual data for each customer including usage start information and one or more items representing customer attributes other than the usage start information, and a determination target time using the customer database A test data generation unit that generates test data that includes individual data of customers who have not started using products or services at the current time, and customers who have started using products or services at the current time Labels with the first label on the individual data and labels with the second label on the individual data of customers who have not started using goods or services A labeling data generation unit that generates a labeling data and calculates a ratio of the number of individual data labeled with the first label in the labeling data as a penetration rate of goods or services, and an item representing customer attributes from the customer A classifier that generates a classifier, which is a rule that determines the probability score that the first label is labeled on the individual data, and a threshold function that is a function of a threshold value of the score with respect to the penetration rate The threshold is calculated by substituting the penetration rate calculated by the labeling data generation unit into the first label for each individual data in the test data from the classifier and each individual data item in the test data. A probability score is defined, and the label for the individual data whose score is equal to or greater than the threshold among the individual data in the test data is the first label Determined, characterized in that the labels for the individual data below score threshold and a determining test data label determining unit that the second label.

  Further, the spread prediction method of the present invention indicates that the use start time of the product or service is indicated when a customer starts using the product or service, and is not used when the product or service is not used. And using a customer database that stores customer data that is a set of individual data for each customer including use start information that represents and one or more items that represent customer attributes other than the use start information. Test data that includes individual data of customers who have not started using the product or service at the specified current time is generated, and the individual data of customers who have started using the product or service at the current time Label one label, label the second label on individual data of customers who have not started using the goods or services, and label the first label Generating learning data that is data obtained by changing the number of different data according to the degree of influence that is a degree that a customer of a product or service urges others to use the product or service within a certain period, Based on the learning data, a classifier that is a rule for determining which one of the first label and the second label is to be labeled on the individual data of the customer from the item representing the attribute of the customer, the classifier and the A label for each individual data in the test data is determined from an item of each individual data in the test data.

  Further, the spread prediction method of the present invention indicates that the customer or service starts when the customer starts using the product or service, and indicates that the product or service is unused when the product or service is not used. Current time for determining the determination target time using a customer database that stores customer data that is a collection of individual data for each customer including use start information and one or more items representing customer attributes other than use start information Generate test data that includes individual data of customers who have not started using products or services in, and label the first data to individual data of customers who have started using products or services at the current time , Generate the labeling data by labeling the second label on the individual data of customers who have not started using the goods or services. This is a rule that calculates the ratio of the number of individual data labeled with a label as the penetration rate of goods or services, and determines the probability score that the first label is labeled on the customer's individual data from the item representing the customer's attribute. A classifier is generated based on the labeling data, and the threshold value is calculated by substituting the penetration rate calculated in the threshold function, which is a function of the score threshold with respect to the penetration rate, for each individual data in the classifier and test data. The probability score that the first label is labeled to each individual data in the test data is determined from the item, and the label for the individual data in which the score is equal to or greater than the threshold among the individual data in the test data is the first label And determining that the label for the individual data having a score lower than the threshold is the second label.

  Further, the spread prediction program of the present invention represents the use start time of the product or service when a customer starts using the product or service, and is unused when the product or service is not used. Popularization installed in a computer having a customer database that stores customer data that is a set of individual data for each customer including use start information that represents customer attributes and one or more items that represent customer attributes other than the use start information A test program for generating test data in the computer using the customer database, the test data being data including individual data of customers who have not started using goods or services at the current time for determining the determination target time Data generation processing, the first data is added to the individual data of customers who have started using the goods or services at the current time. The second label to the individual data of customers who have not started using the product or service, and the number of individual data labeled with the first label is the customer of the product or service within a certain period. Learning data generation processing for generating learning data, which is data that is varied according to the degree of influence that is the degree to which the other person is encouraged to use the product or service, and the individual of the customer from the item representing the attribute of the customer A classifier generating process for generating a classifier, which is a rule for determining whether the data is labeled with a first label or a second label, based on the learning data, and each of the classifier and the test data A test data label determination process for determining a label for each individual data in the test data from the individual data item is performed. To.

  Further, the spread prediction program of the present invention represents the use start time of the product or service when a customer starts using the product or service, and is unused when the product or service is not used. Popularization installed in a computer having a customer database that stores customer data that is a set of individual data for each customer including use start information that represents customer attributes and one or more items that represent customer attributes other than the use start information A test program for generating test data in the computer using the customer database, the test data being data including individual data of customers who have not started using goods or services at the current time for determining the determination target time Data generation processing, the first data is added to the individual data of customers who have started using the goods or services at the current time. Labeling data, generating labeling data in which the second label is labeled on individual data of customers who have not started using the goods or services, and are labeled with the first label in the labeling data A labeling data generation process for calculating a ratio of the number of data as the diffusion rate of the product or service, and a rule for determining a probability score that the first label is labeled on the individual data of the customer from the item representing the attribute of the customer A classifier is generated based on the labeling data, and a threshold is calculated by substituting the penetration rate calculated in the labeling data generation processing into a threshold function that is a function of the threshold value of the score with respect to the penetration rate. A first label for each individual data in the test data from the classifier and each individual data item in the test data. A probability score to be labeled is determined, and among the individual data in the test data, it is determined that the label for the individual data whose score is equal to or greater than the threshold is the first label, and the label for the individual data whose score is less than the threshold A test data label determination process for determining that is a second label is executed.

  Further, the influence degree estimation method of the present invention includes a current time used as an influence degree estimation target time, which is a degree that a customer of a product or service urges others to use the product or service in a certain period. A time interval for designating a predetermined time before the current time and a plurality of temporary influence levels that are candidates for the influence level are input, and a previous time that is a time before the time interval is calculated from the current time. When the customer starts to use the product or service, the use start information represents the use start time of the product or service, and when the product or service is not used, the use start information indicating that the product or service is not used, and the use Using a customer database storing customer data that is a set of individual data for each customer including one or more items representing customer attributes other than start information, the product or service at the current time is stored. Generating the current time data by labeling the first label on the individual data of the customer who has started using the service, and labeling the second label on the individual data of the customer who has not started using the product or service, For each temporary influence degree, the first label is labeled on the individual data of the customer who has started using the product or service at the previous time, and the individual data of the customer who has not started using the product or service The second label is labeled, the previous time data that is the data in which the number of individual data labeled with the first label is changed according to the temporary influence degree is generated, and the attribute of the customer is generated for each previous time data. Based on the previous time, a classifier that is a rule for determining which of the first label and the second label to label the individual data of the customer from the item representing the For each generated classifier, a label to be labeled on the individual data in the current time data is predicted from the classifier and the individual data items in the current time data, and the prediction result and the current time When the error with the data is calculated, the smallest error among the errors calculated for each classifier is specified, and there is one provisional influence corresponding to the smallest error, the provisional influence degree Is determined as an influence degree, and when there are a plurality of temporary influence degrees corresponding to the minimum error, the influence degree is determined based on the plurality of temporary influence degrees.

  The influence estimation program of the present invention represents the use start time of the product or service when the customer starts using the product or service, and is not used when the product or service is not used. Impact of mounting on a computer having a customer database that stores customer data that is a set of individual data for each customer, including usage start information representing the effect and one or more items representing customer attributes other than the usage start information Is a degree estimation program, in which a current time used as an influence degree estimation target time, a time interval for designating a predetermined time before the current time, and a plurality of temporary influence degrees as influence degree candidates are input. , Calculate the previous time that is the time before the time interval from the current time, and start using the goods or services at the current time using the customer database Current time data generation processing for generating current time data by labeling individual data of a customer who has a first label and labeling a second label to individual data of a customer who has not started using the product or service, For each temporary impact level, the first label is labeled on the individual data of the customer who has started using the product or service at the previous time, and the individual data of the customer who has not started using the product or service is The previous time data group generation processing for generating the previous time data which is the data obtained by changing the number of individual data labeled with two labels and the first label according to the temporary influence degree, for each previous time data , A classifier that is a rule for determining which of the customer's individual data is labeled with the first label or the second label from the item representing the customer's attribute at the previous time For each individual classifier generated for each previous time data, for each classifier generated for each previous time data, from the classifier and each individual data item in the current time data, An error group calculation process for calculating an error between a prediction result and the current time data is predicted, and a minimum error is specified among errors calculated for each classifier. When there is one temporary influence degree corresponding to the minimum error, the temporary influence degree is determined as the influence degree. When there are a plurality of temporary influence degrees corresponding to the minimum error, the plurality of temporary influence degrees are determined. An influence degree calculation process for determining the influence degree based on the temporary influence degree is executed.

  ADVANTAGE OF THE INVENTION According to this invention, the time when a customer starts using goods and service can be estimated for every customer. Moreover, according to this invention, the influence degree of goods or a service can be estimated according to time.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  Not only those who use products and services for a fee, but also those who use products and services for free are called customers.

  In addition, the influence degree estimation system and the spread prediction system described below perform influence degree estimation and spread prediction for one product or service.

Embodiment 1. FIG.
FIG. 1 is a block diagram showing an example of an influence estimation system of the present invention. The influence degree estimation system of the present invention includes an influence degree estimation device 3 that estimates the degree of influence (the degree that a customer of a product or service urges others to use the product or service within a certain period), and a customer database. (Hereinafter referred to as customer DB) 2. The concept of products includes not only finished products such as products but also the minimum unit for maintaining the functions of products.

  The customer DB 2 is storage means for storing customer data and can be called customer data storage means. Customer data is a set of data determined for each individual customer. Hereinafter, data for each individual customer is called individual data. In the customer DB 2, the individual data for each customer includes usage start information and one or more items representing customer attributes other than the usage start information. When the customer starts using the product or service, the usage start information indicates the time when the customer started using the product or service, and the customer has not used the product or service (that is, the customer has not started using the product or service yet). If not, it is information indicating that it is not used. As items representing customer attributes other than use start information, for example, characteristics such as gender, age, address area classification, work area area, use history of other products, and various preferences can be used. However, it is not limited to these. Each individual information includes one or more of these items. Items representing customer attributes can be collected by various methods such as, for example, input by a store clerk at a store, registration at the time of starting use of a product, customer membership card, questionnaire, and the like.

  The influence level estimation device 3 includes a current time data generation unit 31, a previous time data group generation unit 32, a previous time classifier group generation unit 33, an error group calculation unit 34, and an influence level calculation unit 35.

  The current time data group generation unit 31 receives the current time, a time interval for designating a certain time before the current time, and a plurality of temporary influence levels. As will be described later, the current time, the time interval, and a plurality of temporary influences may be input to the current time data group generation unit 31 via an input device such as a keyboard. The current time in this embodiment is a time used as an influence degree estimation target time (impact degree estimation target time), and a time interval is a period in which the influence degree is to be estimated. In the present embodiment, an influence degree in a period from a time pointed back by a time interval from the time point to the influence degree estimation target time is estimated using a past time point as the influence degree estimation target time. Accordingly, the time at a certain point in the past is input as the current time. The time before the time interval from the current time (the time that is back by the time interval from the current time) is described as the previous time. The current time data group generation unit 31 calculates the previous time by subtracting the time interval from the current time.

  The input temporary influence degree is a candidate of the influence degree to be obtained, and the influence degree estimation device 3 obtains the influence degree from a plurality of temporary influence degrees. The numerical value input to the current time data group generation unit 31 as the temporary influence degree is a numerical value of 0 or more. For example, the user of the influence degree estimation system selects a plurality of influence degree candidates from values ranging from 0 to about 50, and inputs them as temporary influence degrees. If a large number of provisional influence degrees are input as candidates, the influence degree estimation accuracy is increased.

  The current time data group generation unit 31 generates one current time data based on the current time and the customer DB 2. For the current time data, the first label is labeled on the individual data of the customer who has started using the product or service at the current time, and the second is added to the individual data of the customer who has not started using the product or service at the current time. This is data with labels. The first label is information indicating a state where the use of the product or service is started, and the second label is information indicating a state where the use of the product or service is not started. Hereinafter, the first label is described as positive (or +), and the second label is described as negative (or-). The current time data group generation unit 31 generates current time data using the customer DB 2 (that is, based on customer data stored in the customer data DB 2).

  The previous time data group generation unit 32 generates a plurality of previous time data based on the previous time, the plurality of temporary influence degrees, and the customer data stored in the customer DB 2. For the previous time data, the first label (correct) is labeled on the individual data of the customer who has started using the product or service at the previous time, and the individual data of the customer who has not started using the product or service at the previous time The second label (negative) is labeled and data is weighted according to the temporary influence degree. The weighting means that the number of individual data labeled with the first label is changed relative to the number of individual data labeled with the second label. The previous time data group generation unit 32 generates previous time data for each temporary influence degree. Since there are a plurality of temporary influence degrees, a plurality of previous time data are generated.

  The previous time classifier group generation unit 33 generates one classifier for each previous time data generated by the previous time data group generation unit 32. Therefore, the previous time classifier group generation unit 33 generates a plurality of classifiers. The classifier is a rule that determines whether positive (first label) or negative (second label) is to be labeled on individual customer data from items representing customer attributes. In other words, it is a rule that defines an item representing a customer attribute as an independent variable, a possible value of the dependent variable as positive or negative, and determines the dependent variable from the independent variable.

  The error group calculation unit 34 predicts the label of the current time data using each classifier generated by the previous time classifier group generation unit 33. That is, the error group calculation unit 34 compares the classifier with each individual data item (item indicating the customer's attribute) in the current time data, and converts each individual data in the current time data into the individual data. Predict labels to be labeled. Further, each individual data in the current time data has already been actually labeled. The error group calculator 34 calculates an error between the label predicted from the classifier and the item and the label actually labeled with the current time data. The error group calculation unit 34 performs this process for each classifier. Therefore, the error group calculation unit 34 calculates a plurality of errors.

  The influence degree calculation unit 35 specifies the minimum error among the errors calculated for each classifier, and determines the provisional influence degree corresponding to the minimum error as the influence degree. Specifically, when the number of provisional influence levels corresponding to the minimum error is one, the influence degree calculation unit 35 determines the provisional influence degree as an influence degree. In addition, when there are a plurality of provisional influence degrees corresponding to the minimum error, the influence degree is determined based on the plurality of provisional influence degrees. For example, when there are a plurality of provisional influence degrees corresponding to the minimum error, an average value of the plurality of provisional influence degrees is calculated, and the average value is determined as the influence degree. Hereinafter, a case where there are a plurality of provisional influence levels corresponding to the minimum error and an average value of the provisional influence degrees is defined as the influence degree will be described as an example.

  The current time data generation unit 31, the previous time data group generation unit 32, the previous time classifier group generation unit 33, the error group calculation unit 34, and the influence level calculation unit 35 included in the influence degree estimation device 3 are, for example, influence degree estimation. It is realized by a CPU that operates according to a program. That is, the CPU that reads the influence degree estimation program from the storage device provided in the influence degree estimation system is the current time data generation unit 31, the previous time data group generation unit 32, the previous time classifier group generation unit 33, and the error group calculation unit. 34 and the influence calculation unit 35 may be operated.

  Note that the current time, the time interval, and a plurality of temporary influence levels may be input to the current time data group generation unit 31 via an input device such as a keyboard. Further, an output device for outputting the influence degree obtained by the influence degree calculation unit 35 may be provided. FIG. 2 is a block diagram illustrating an example of an influence degree estimation system including an input device and an output device. Constituent elements similar to those shown in FIG. 1 are denoted by the same reference numerals as those in FIG.

  The input device 1 is an input device such as a keyboard. The current time, the time interval, and a plurality of temporary influences may be input via such an input device 1. However, the current time, the time interval, and a plurality of temporary influences may be input to the current time data group generation unit 31 in other manners.

  The output device 4 is an output device such as a display device. The influence degree calculation unit 35 may output the obtained influence degree to the output device 4. For example, when the output device 4 is a display device, the influence degree may be displayed on the output device 4. Further, the output device 4 may be a printing device.

Next, the operation will be described.
FIG. 3 is a flowchart showing an example of processing progress of the influence degree estimation system of the present invention. For example, when the current time, time interval, and a plurality of temporary influence levels are input to the current time data generation unit 31 via the input device 1 (see FIG. 2), the influence level estimation system operates as follows.

  First, the current time data generation unit 31 calculates the previous time and creates current time data (step A1). Further, in step A1, the current time data generation unit 31 sets an initial value of a variable (referred to as ErrorMin) indicating the minimum value of errors calculated by the error group calculation unit 34, and the variables i and p are set. Thus, initial values of i = 1 and p = 0 are set. The initial value of ErrorMin may be a maximum value that can be taken by the error calculated by the error group calculation unit 34 or a value that is sufficiently larger than a value that can be taken by the error. The variable i is a variable for sequentially specifying a plurality of input temporary influence degrees. For example, if i = 1, it means that the first temporary influence degree is designated. P is a variable for designating a temporary influence level corresponding to the minimum error. For example, if there are p temporary influences corresponding to the minimum error, the first to pth temporary influences corresponding to the minimum error are respectively Imp [1],..., Imp [p]. And

  In step A1, the current time data generation unit 31 sets the previous time as the current time-time interval. That is, the time obtained by subtracting the time interval from the current time is set as the previous time.

  In step A1, the current time data generation unit 31 reads customer data stored in the customer DB2. Then, individual data of customers who start using the product or service at the current time (individual data whose use start information indicates the time before the current time) is labeled positive and the use of the product or service at the current time The current time data is generated by labeling the individual data of customers who have not started the process negative.

  At this time, the current time data generation unit 31 includes, among the individual data belonging to the customer data stored in the customer DB 2, the individual data excluding the individual data of the customer who started using the product or service at the previous time. Current time data may be generated from the set. That is, the individual data whose use start information is before the previous time may be excluded, and the remaining individual data may be labeled positively or negatively. Hereinafter, the case where the individual data whose use start information is the time before the previous time is excluded so as not to be included in the current time data will be described as an example. Each individual data belonging to customer data may be labeled positively or negatively.

  The operation | movement which produces | generates present time data from customer data is shown concretely. FIG. 4 is an explanatory diagram illustrating an example of customer data stored in the customer DB 2. Each row shown in FIG. 4 is individual data of one customer, and a set of the individual data is customer data. The example shown in FIG. 4 shows a case where identification information (ID) of each individual data is given. Although FIG. 4 illustrates the case where the month when the customer starts using the product or service is used as the use start time, the use start time may be expressed in units of days or hours. In addition, when the use start information is represented by “?”, The “?” Means “unused (that is, the state where the use has not started)”, but other symbols “unused”. May be represented. FIG. 4 illustrates a case where each individual data includes N items representing customer attributes other than the use start information. In this example, item 1 is gender. For each item, if it is unknown, it can be represented by a symbol indicating a missing value. In this example, “?” Is used as a symbol representing a missing value.

  It is assumed that the current time is April 2007 and the calculated previous time is February 2007. The current time data generation unit 31 leaves the item values of the respective items as they are, eliminates the use start information, creates a class indicating whether or not the use is started instead, and labels the individual data as positive or negative. It should be noted that male / female may be converted to a numerical value such as 0, 1 if necessary in a later step of creating a classifier. Also, the current time data generation unit 31 includes individual data (ID = 1 individual data) whose usage start information is before the previous time (February 2007) among the individual data in the customer data shown in FIG. Data) and other individual data are labeled positive if the product or service has begun to be used at the current time (April 2007), otherwise it is labeled negative. For example, in the individual data with ID = 2, since the use has started from March 2007, it is assumed to be positive. Further, the individual data with ID = 3 has been used since May 2007, and since the use of goods or services has not started in April 2007, it is negative. In addition, the individual data whose use start information is “? (Unused)” is also negative. FIG. 5 is an explanatory view showing an example of the current time data generated from the customer data shown in FIG. Customer data (ID = 1) of a customer who has started use from January 2007 before the previous time is excluded from the current time data, and customer data (ID = 2) of customers who have started use before April 2007 is excluded. ) Is labeled positive (+) and other customer data is labeled negative (-). Although FIG. 5 shows the current time data in which the ID is left for ease of understanding, items unrelated to the classification may be deleted in advance.

  After step A1, the previous time data group generation unit 32 generates previous time data corresponding to the temporary influence degree False [i] using the i-th temporary influence degree (hereinafter referred to as False [i]). (Step A2). The previous time data group generation unit 32 reads the customer data stored in the customer DB 2, labels the individual data of the customers who have started using the product or service before the previous time, and labels the product or service at the previous time. Label negative data for individual customers who have not started using. That is, individual data whose use start information (see FIG. 4) is a time before the previous time is labeled positive, and individual data whose use start information is not before the previous time is labeled negative. At this time, weighting according to the temporary influence degree [i] is not performed. FIG. 6 is an explanatory diagram illustrating an example of data before weighting is performed in the previous time data generation process. In the individual data with ID = 1, the use is started from January 2007, and the use is started at the previous time (February 2007). Therefore, positive (+) is labeled. In addition, in the individual data after ID = 2, the use start time is later than the previous time (February 2007), and the use has not started at the previous time. Both labels are negative (−) (see FIGS. 4 and 6). Note that individual data is not excluded when generating previous time data.

  Then, the previous time data group generation unit 32 performs weighting according to the temporary influence degree False [i]. That is, according to the temporary influence degree False [i], the number of positive individual data is changed relative to the number of negative individual data. In this example, the function value of the function of the temporary influence degree False [i] is calculated, and the case where the individual data labeled positive is increased to the function value multiple is taken as an example.

  As a function of the provisional influence level False [i], for example, the ratio of the number of customers who have started using the product or service at the previous time and the number of customers who have not started using the product or service at the previous time You may use the function used as a coefficient of a temporary influence degree. That is, the functions exemplified below may be used.

  False [i] x (number of unused users at the previous time) / (number of users at the previous time)

  In the above function, the “number of unused users at the previous time” is the number of customers who have not yet started using the product or service at the previous time. That is, in the customer data, the number of individual data whose usage start information is not before the previous time. Further, the “number of users at the previous time” is the number of customers who have started using the product or service at the previous time. That is, in the customer data, the number of individual data whose use start information is before the previous time. In this way, a function may be used in which the ratio obtained by dividing the “number of unused users at the previous time” by the “number of users at the previous time” is the coefficient of False [i].

  Further, as a function of the temporary influence degree False [i], for example, a function using a ratio between the number of customers who have started using the product or service at the previous time and the total number of customers as a coefficient of the temporary influence degree may be used. Good. That is, the functions exemplified below may be used.

    False [i] x (number of customers) / (number of users at the previous time)

  In the above function, the “number of customers” is the total number of customers (the total number of individual data in the customer data). The number of customers who have started using the previous time and the number of customers who have already started using the previous time. There is no sum of the number of customers. The “number of users at the previous time” is the same as in the case of the above function. In this way, a function may be used in which a ratio obtained by dividing “number of customers” by “number of users at the previous time” is a coefficient of False [i].

  The previous time data group generation unit 32 increases the individual data labeled positive as the function value multiple of such a function. For example, it is assumed that the function value is 5 when the temporary influence degree is False [i]. In this case, a copy of the individual data labeled positive may be generated so that the number of individual data labeled positive is five times. The data obtained as a result is the previous time data. FIG. 7 is an explanatory diagram illustrating an example of previous time data obtained by weighting. In the example shown in FIG. 6, the individual data with ID = 1 is positive, and the other individual data is negative. Therefore, the previous time data group generation unit 32 generates five pieces of data having the same contents as the individual data with ID = 1. However, the IDs are distinguished as 1 ', 1 ", etc., respectively.

  If only a copy of the individual data labeled positive is generated, the individual data labeled positive may be weighted, and then the other individual data may be negatively labeled.

  In addition, the value of the coefficient of the temporary influence degree in the two exemplified functions decreases as time passes and the number of users increases. The coefficient of the temporary influence degree is not limited to such a coefficient. Further, the value of the temporary influence degree False [i] may be used as a function value as it is. In this case, the coefficient is 1.

  In the above example, the case where the individual data labeled positive is increased by a multiple of the function value, but in weighting, the number of individual data labeled positive is changed to the number of individual data labeled negative. What is necessary is just to change relatively with respect to it. For example, if the weight (function value) is 5 and the number of individual data labeled as positive is 10 times, the number of individual data labeled as negative (the number of unused users) may be doubled. Further, if the weight is 0.5 and the number of individual data labeled positive is to be doubled, the number of individual data labeled negative is to be doubled. Here, an example is given in which the number of individual data is increased by 10 times. However, when conditions such as increasing the number of individual data by k times are determined in this way, the k is input from the outside in advance. Just keep it.

  Here, the individual data is actually increased by creating a copy of the individual data. However, when the classifier is created in the next step, the function value of the temporary influence level False [i] is used as the weighting parameter. It may be used as That is, it is possible to determine how many times the individual data labeled as positive is increased, and generate a classifier using the result as a parameter.

  After the previous time data group generation unit 32 generates the previous time data corresponding to False [i] in step A2, the previous time classifier group generation unit 33 uses the previous time data to correspond to the previous time data. A classifier (denoted as previous time classifier i) is generated (step A3). The ID is not necessary for creating the classifier, and the previous time classifier group generation unit 33 may delete the ID. The previous time classifier group generation unit 33 may generate a classifier by various data mining techniques. For example, the classifier may be generated by any one or a combination of methods such as multiple regression analysis, decision tree, neural network, support vector machine, and Bayesian network.

  Hereinafter, a specific example of the classifier will be described by taking a case where a decision tree is used as the classifier as an example. In this example, “whether or not to play golf” is a dependent variable, and “weather”, “weather”, “humidity”, and “whether wind is strong” are independent variables. FIG. 8 is an explanatory diagram showing an example of a combination of known independent variables and dependent variables. FIG. 9 shows an example of a classifier, where a decision tree is used as a classifier. FIG. 10 is an explanatory diagram illustrating an example of a dependent variable predicted from a classifier. As shown in FIG. 8, if there is a combination of a known independent variable and dependent variable, a decision tree can be generated from the known independent variable and dependent variable. The dependent variable can be predicted from the decision tree and the known independent variable. For example, if the decision tree shown in FIG. 9 is obtained, the dependent variable exemplified in FIG. 10 can be predicted from the independent variable exemplified in FIG. Branching according to the item value of an item is called division.

  In the example shown in FIG. 8, each row corresponds to individual data labeled “go / do not play golf”. Hereinafter, each row shown in FIG. 8 will be described as individual data. Further, “go golf” corresponds to a positive (+) label, and “do not play golf” corresponds to a negative (−) label. In the decision tree, information that summarizes the number of individual data for each label that can be taken by the dependent variable (for example, for each label of “Yes (positive)” and “No (negative)”) is used as a node. For example, in the node of the route shown in FIG. 8, the information “Yes: 9, No: 5” is used as the node.

  When the previous time data is given, the previous time classifier group generation unit 33 determines which item is first divided. At this time, the previous time classifier group generation unit 33 calculates an evaluation value at the time of division for each of items 1 to N, and selects an item having the largest evaluation value as an item most suitable for division. Here, although the case where the difference between the entropy of the node before the division and the entropy after the division is used as an evaluation value is taken as an example, the evaluation value may be obtained by another calculation method. The entropy of a node is -qlogq- (1-q) log (1-), where q is the ratio of individual data with a positive (+) label and 1-q is the ratio of individual data with a negative (-) label. q). The entropy of the node after the division is a weighted average of the entropy of each node after the division.

  For example, in the previous time data, if “positive” has 9 data and “negative” has 5 data, the root node is “positive: 9, negative: 5”. In this case, since there are nine positive (+) data and five negative (-) data, the entropy of the root node is-(9/14) * log (9/14)-(5 / 14) × (5/14) = 0.940.

  The previous time classifier group generation unit 33 obtains a node obtained by dividing the root node by one item. That is, information (node) representing the number of positive and negative individual data is generated for each item value of the item. For example, when the possible value of the item 1 is “0” or “1”, and the value of the item 1 is “0”, it is assumed that there are 5 positive data and 2 negative data, and the value of the item 1 is “ In the case of “1”, it is assumed that there is 0 data for positive and 7 data for negative. In this case, a node “positive: 5, negative: 2” and a node “positive: 0, negative: 7” are generated as nodes that branch depending on the value of item 1 being “0” or “1”. . The previous time classifier group generation unit 33 calculates the entropy of each node after division, and obtains a weighted average of entropy of each node according to the number of individual data counted as positive or negative in each node after division. In the above example, since the total number of individual data is 7 in both the nodes “positive: 5, negative: 2” and “positive: 0, negative: 7”, the weighting coefficient when performing the weighted average is Each node is (7/14). Therefore, in this example, the previous time classifier group generation unit 33 calculates the entropy after the division as follows.

(7/14) × {− (5/7) × log (5/7) − (2/7) × log (2/7)}
+ (7/14) × {− (0/7) × log (0/7) − (7/7) × log (7/7)}
= 0.432

  Therefore, in the case of this example, the previous time classifier group generation unit 33 calculates the evaluation value when divided by item 1 as 0.940−0.432 = 0.508.

  The previous time classifier group generation unit 33 calculates not only the item 1 but also other items in the same manner, calculates an evaluation value when the item is divided, and determines that the item is divided by the item having the maximum evaluation value. In this way, the item to divide the root node is determined.

  In the above item 1 example, the case where the value that the item 1 can take is only “0” or “1” is shown. When the item value is age and the value is a continuous value such as 20, 21, 22, it is also determined where the item value is divided. In this case, the previous time classifier group generation unit 33 uses the intermediate value between the item values as a threshold value, and for each threshold value, “below the threshold value” and “greater than the threshold value” The evaluation value in the case of dividing into two is obtained. Then, by selecting a case where the evaluation value is maximized, it is also determined which item value is to be divided. For example, in the case where the item values are continuous with 20, 21, 22,..., The evaluation values when divided by “20.5 or less” and “greater than 20.5”, “21.5 or less” and “ An evaluation value or the like when dividing by “greater than 21.5” may be calculated and divided so that the evaluation value becomes the highest.

  The previous time classifier group generation unit 33 performs the same process as described above for each node after division, and sequentially repeats the process of determining which item is to be divided next. Further, the previous time classifier group generation unit 33 stops dividing the node when a predetermined condition is satisfied. The predetermined condition is, for example, a condition that “the classes of individual data in the node are all the same” or “the number of individual data counted as positive or negative in the node is equal to or less than a predetermined number (for example, 2)”. Conditions may be used. When the former condition is adopted, when all the individual data in the node becomes positive or negative, the division of the node is not continued. As described above, the previous time classifier group generation unit 33 sequentially repeats the division from the root node to generate a tree-structured decision tree.

  Further, as described above, the previous time classifier group generation unit 33 generates a tree-structured decision tree, and then performs pruning on the decision tree. In the decision tree, the final node generated by dividing is called a leaf. Assume that the number of data classified into a certain leaf is N (that is, the number of individual data counted as positive or negative is N). It is assumed that E data is erroneous in N data classified into these leaves. Under this assumption, it is considered that the event of error is observed E times during N trials, and can be considered as a binomial distribution in which the probability of the event of error occurring in a sample of size N is r. If the upper limit of r is expressed as U_CF (E, N) with respect to the reliability CF given in advance, the expected value at which an error occurs in N data is N × U_CF (E, N). The previous time classifier group generation unit 33 generates an expected value of an error in the parent (expected value in which an error occurs) and an expected value of an error in the child leaf for the parent node whose child nodes are all leaves. Compare the total. If the sum of the expected values at the child is larger than the expected value of the parent error, the previous time classifier group generation unit 33 degenerates the leaf and sets the parent as the leaf. The previous time classifier group generation unit 33 sequentially repeats this process to prun the leaves of the entire decision tree.

  When degenerating a leaf, the previous time classifier group generation unit 33 may delete the leaf and use the parent node of the deleted leaf as the leaf. For example, in the decision tree illustrated in FIG. 9, when the node divided according to the value of the item “humidity” is degenerated, the previous time classifier group generation unit 33 has the humidity item value of 70% or less. Delete the node “Yes: 2, No: 0” representing the number of individual data and the node “Yes: 0, No: 3” representing the number of individual data whose humidity value is higher than 70%. The parent node “Yes: 2, No: 3” of the two nodes may be regarded as a leaf.

  When generating a decision tree as a classifier, for example, a decision tree may be generated by determining and pruning a decision tree as described above.

  When the classifier (previous time classifier i) is generated from the previous time data corresponding to False [i] in step A3, the error group calculation unit 34 uses the previous time classifier i to label the current time data. Predict (step A4). That is, for each individual data in the current time data, the error group calculation unit 34 collates an item indicating the customer attribute in the individual data with the previous time classifier i, and labels the individual data to be labeled Predict.

  For example, when the classifier is a decision tree, the error group calculation unit 34 refers to the item value of the individual data in the current time data for the item of the node of the route, and traces the child node according to the item value. If the error group calculation unit 34 has traced to the leaf node, the error group calculation unit 34 may use the label with the larger count number at the leaf node as the prediction result. For example, if “positive: 3, negative: 0” in a leaf node, “positive” may be predicted.

  After step A4, the error group calculation unit 34 calculates an error between the prediction result of the label labeled on each individual data in the current time data and the current time data (step A5). That is, since each individual data in the current time data is labeled with a positive or negative label when the current time data is generated, it is labeled with the prediction result in step A4 and the actual current time data. Predict the error from the label. This error is referred to as Err [i].

  In step A5, for example, the error group calculation unit 34 compares the label predicted for each individual data in the current time data with the label actually labeled on each individual data in the current time data. The number of different individual data may be counted, and the count value may be obtained as Err [i].

  Alternatively, the error group calculation unit 34 may label the number of individual data predicted to be positive among the labels predicted for each individual data in the current time data, and actually label the positive in the current time data. A difference from the number of obtained individual data may be calculated, and the difference may be obtained as Err [i]. That is, the absolute value of the difference between the predicted number of positive labels and the number of individual data (actual number of users) actually labeled positive in the current time data may be Err [i].

  After step A5, the influence calculation unit 35 compares the error Err [i] with the minimum error value ErrorMin, and determines whether Err [i] is less than ErrorMin (step A6). If Err [i] is less than ErrorMin (Yes in Step A6), the influence calculation unit 35 initializes p to 0 and substitutes Err [i] for ErrorMin (Step A11). If Err [i] is less than ErrorMin, it means that a minimum value smaller than the error that has been minimized so far has been found. In this case, in step A11, ErrorMin is updated with the minimum value, and when there are a plurality of temporary influence degrees corresponding to the error, a variable p for individually specifying each temporary influence degree is initialized.

  After step A11, the influence calculation unit 35 increments the value of p by 1, and substitutes False [i] for Imp [p] (step A12). In Step A6, when it is determined that Err [i] is equal to or greater than ErrorMin (No in Step A6), the influence calculation unit 35 determines whether Err [i] is equal to ErrorMin (Step A7). ). Also when it determines with it being equal here (Yes in step A7), the influence calculation part 35 increments the value of p by 1, and substitutes False [i] to Imp [p] (step A12).

  When the process proceeds from step A11 to step A12, p = 1, and the temporary influence degree False [i] currently focused on is set as the first temporary influence degree that minimizes the error. In Step A7, when Err [i] corresponding to the focused False [i] is equal to ErrorMin and the process proceeds to Step A12, p becomes a value of 2 or more. In this case, at least one temporary influence degree corresponding to the minimum error has already been found, and the influence degree calculation unit 35 determines the temporary influence degree False [i] currently focused on as the p-th error with the minimum error. The provisional influence level is determined.

  After step A12 or when it is determined in step A7 that Err [i] is not equal to ErrorMin (No in step A7), the influence degree calculation unit 35 increments the value of i (step A8). Next, the influence degree calculation unit 35 determines whether or not the value of i is equal to or less than the number of temporary influence degrees (M) that is input first (step A9). If i is less than or equal to the temporary influence frequency M (Yes in step A9), the processes after step A2 are performed on the temporary influence degree False [i] determined by i incremented in step A8. That is, the temporary influence degree to which attention is paid is changed, and the processes after step A2 are performed.

  If the value of i is larger than the temporary influence frequency M (No in step A9), the process proceeds to step A10. Here, i represents the number of temporary influences for which error calculation has been completed. In step A10, the influence degree calculation unit 35 calculates an average value of each temporary influence degree corresponding to the minimum error, and outputs the average value as the influence degree (step A10). If there is only one temporary influence corresponding to the minimum error, p = 1. At this time, the influence calculation unit 35 may output Imp [1] as the influence. In addition, when there are a plurality of provisional influence levels corresponding to the minimum error, p is a value of 2 or more. At this time, the influence degree calculation unit 35 may calculate an average value of the influence degrees from Imp [1] to Imp [p] and output the average value as the influence degree.

  For example, the influence degree calculation unit 35 displays the influence degree on the output device 4. Or you may output an influence degree in another output mode. For example, the output device 4 may be a printing device, and the output device 4 may print the influence degree.

Next, the effect of the first embodiment will be described.
In the first embodiment, a prediction error is obtained for each of a plurality of inputted temporary influence degrees, and an average of the temporary influence degrees that minimizes the error is set as the influence degree. Therefore, it is possible to estimate the degree of influence in the period from the previous time to the current time. In addition, by determining the input current time and time interval in accordance with the period for which the degree of influence is to be examined, the degree of influence in the period to be examined can be estimated.

  Further, the error group calculation unit 34 compares the predicted label with the label actually labeled on each individual data of the current time data, counts the number of individual data that are different from each other, and the count value Is the error (Err [i]), it is possible to improve the estimation performance of the influence degree.

  In addition, among the labels predicted by the error group calculation unit 34 for each individual data in the current time data, the number of individual data predicted to be positive and the actual positive label in the current time data are labeled. When the difference (Err [i]) is the difference from the number of obtained individual data, the customer who should start using at the current time should start at the next time or start using at the next time It is possible to estimate the degree of influence by smoothing out fine fluctuations such as the customer starting to use the service at the current time early.

  Further, in the above-described embodiment, the case where the influence level at one input current time is estimated has been described. However, a plurality of types of current time are input, and the processing after step A1 is performed for each current time. And the degree of influence may be obtained for each current time. In this case, it is possible to estimate the degree of influence that changes every moment.

  Further, for example, when a plurality of current times are input and an influence degree is obtained for each current time, the following operation may be performed. That is, when it is specified by the user of the influence degree estimation system to obtain the influence degree for each current time, the process after step A1 is performed for each current time, and the influence degree for each current time is obtained. The influence degree calculation unit 35 may obtain the average value of the influence degree at each current time as the influence degree. If the degree of influence is calculated in this way, if it seems that the estimation error of the degree of influence can be reduced if many times are used in a period that is not easily affected by the advertisement / promotion activities, the degree of influence with a small estimation error should be obtained. Can do. In addition, when calculating the average value of the degree of influence at each current time, the closer to the most recent (that is, the degree of influence that the current time is late) is weighted, the weighted average value is obtained, The weighted average value may be used as the influence degree.

  When the user of the influence level estimation system designates a plurality of current times, the current time minimum value, the maximum value, and an interval for changing the current time may be input to the current time data generation unit 31. When the current time minimum value, the maximum value, and the interval for changing the current time are input, the current time data generation unit 31 divides the period from the minimum value to the maximum value of the current time for each interval, In addition to the input minimum value and maximum value, the time at the boundary may be the current time. Then, the degree of influence may be obtained for each current time, and the moving average or the moving weighted average may be obtained.

  Alternatively, a plurality of current times may be input, an error in each temporary influence degree may be calculated for each current time, and the influence degree may be determined from the error in each temporary influence degree at each current time. For example, assume that a plurality of current times T1 to Tx are input. In this example, the influence degree estimation system performs steps A1 to A5 with respect to the current time T1 to obtain Err [1]. After step A5, the error calculation unit 35 performs steps A8 and A9, and sequentially determines errors Err [2] to Err [M] when i = 2 or more. Similarly, Err [1] to Err [M] are obtained for other current times. FIG. 11 shows the error in each temporary influence degree obtained at each current time in this way. After obtaining Err [1] to Err [M] at each current time, the error calculator 35 calculates the sum of Err [i] at each current time for each i from 1 to M, and the sum. The temporary influence degree determined by i that minimizes may be used as the influence degree. In the example illustrated in FIG. 11, the sum of Err [i] is obtained for each row, and False [i] determined by i that minimizes the sum may be used as the degree of influence.

  In addition, when the user of the influence degree estimation system designates a plurality of temporary influence degrees, the maximum value and the minimum value of the temporary influence degree and an interval for changing the temporary influence degree may be input to the current time data generation unit 31. . When the current time data generation unit 31 receives the maximum value, the minimum value, and the interval for changing the temporary influence degree, the range of the value from the minimum value to the maximum value of the temporary influence degree is set to the period. In addition to the input minimum value and maximum value, the boundary value may be determined as the temporary influence degree.

  Further, in the first embodiment, the influence degree is described to be output. However, in order for the user to know the reliability of the influence degree, a calculated error or the like may be output. For example, the influence degree calculation means 35 may cause Err [i] to be output to the output device 4 in addition to the influence degree.

  Further, in the first embodiment, the case where the influence degree estimation system includes the customer DB 2 has been described. However, the customer DB 2 is provided outside the influence degree estimation system, and the influence degree estimation system performs processing with reference to the customer DB 2. The structure which performs this may be sufficient.

Embodiment 2. FIG.
FIG. 12 is a block diagram showing an example of a spread prediction system of the present invention. Components similar to those in the first embodiment are denoted by the same reference numerals as those in FIG. The spread prediction system of the present invention includes a spread prediction device 5 that determines whether or not to start using a product or service for each customer at a predicted time, and a customer database (hereinafter referred to as customer DB) 2. With.

  The customer DB 2 of this embodiment is a storage unit that stores customer data, and is the same as the customer DB 2 (see FIG. 1) in the first embodiment. The customer data stored in the customer DB 2 of the present embodiment is the same as the customer data in the first embodiment. That is, the customer data is a set of individual data defined for each individual customer as shown in FIG. In the customer DB 2, each individual data includes usage start information and one or more items representing customer attributes other than the usage start information. When the customer starts using the product or service, the use start information is the time when the customer starts using the product or service. Further, when the customer does not use the product or service, the fact that it is not used is the use start information.

  The spread prediction device 5 receives the current time and the influence level. The current time in the second embodiment defines a determination target time for determining whether or not the customer has started using the product or service. The influence degree is preferably the influence degree obtained by the influence degree estimation system of the first embodiment, but may be the influence degree determined by the user of the popularization prediction system. In addition, the current time in the present embodiment may be a future time.

  The spread prediction device 5 includes a test data generation unit 51, a learning data generation unit 52, a classifier generation unit 53, and a test data label determination unit 54.

  The test data generation unit 51 generates test data including individual data of customers who have not started use at the current time based on the current time and the customer DB 2.

  The test data is data including individual data of customers who have not started using goods or services at the current time. The test data preferably includes only individual data of customers who have not started using goods or services at the current time, but includes individual data of customers who have started using goods or services at the current time. Also good. Hereinafter, a case where the test data includes only individual data of customers who have not started using the product or service at the current time will be described as an example.

  The learning data generation unit 52 generates learning data based on the input current time and degree of influence and customer data stored in the customer DB 2. For learning data, the first label is labeled on the individual data of the customer who has started using the product or service at the current time, and the second label is applied on the individual data of the customer who has not started using the product or service at the current time. The data is labeled and weighted according to the input degree of influence. As described in the first embodiment, the weighting means that the number of individual data labeled with the first label is changed relative to the number of individual data labeled with the second label. Similarly to the first embodiment, the first label is described as positive (or +), and the second label is described as negative (or −).

  The classifier generation unit 53 generates a classifier using the learning data generated by the learning data generation unit 52.

  The test data label determination unit 54 predicts the label of each individual data in the test data by the classifier. That is, each individual data item in the test data is collated with the classifier to determine whether the label is positive or negative. The customer of the individual data determined to be positive is predicted to start using the product or service at the time after the current time. The customer of the individual data determined to be negative is predicted not to start using the goods or services yet at the time after the current time. The time next to the current time is a time obtained by adding a predetermined fixed time to the current time.

  The test data label determination unit 54 may cause the output device (not shown in FIG. 12) to output a label prediction result (determination result) for each individual data in the test data. For example, the prediction result may be displayed on a display device or the like, or may be printed on a printing device or the like.

  In addition, the test data generation unit 51 and the learning data generation unit 52 may receive the current time from the outside. Further, the learning data generation unit 52 may receive an influence degree from the outside. At this time, for example, the current time and the degree of influence may be input via an input device (not shown in FIG. 12) such as a keyboard. Alternatively, when the influence degree estimation system according to the first embodiment obtains the influence degree at a certain current time, for example, the influence degree calculation means 35 of the influence degree estimation system according to the first embodiment generates test data for the current time. The degree of influence may be passed to the learning data generation unit 52.

  The test data generation unit 51, the learning data generation unit 52, the classifier generation unit 53, and the test data label determination unit 54 included in the spread prediction device 5 are realized by a CPU that operates according to a spread prediction program, for example. That is, the CPU that has read the diffusion prediction program from the storage device provided in the diffusion prediction device 5 operates as the test data generation unit 51, the learning data generation unit 52, the classifier generation unit 53, and the test data label determination unit 54. Also good.

Next, the operation of the spread prediction system will be described.
FIG. 13 is a flowchart showing an example of processing progress of the spread prediction system of the present invention. For example, when the current time and the degree of influence are input to the spread prediction device 5 via an input device (not shown in FIG. 13) such as a keyboard, the spread prediction system operates as follows.

  First, the test data generation unit 51 generates test data (step B1). The test data generation unit 51 extracts only individual data of customers who have not started using the product or service from the customer data stored in the customer data DB 2 and tests the set of the individual data. Data can be used. Specifically, it is only necessary to extract only individual data whose use start information (see FIG. 4) is not before the current time and use the set of the individual data as test data. In this example, the test data includes only individual data of customers who have not started use at the current time. However, individual data of customers who have started using at the current time is included in the test data. It may be.

  After step B1, the learning data generation unit 52 generates learning data (step B2). The learning data generation unit 52 reads customer data stored in the customer DB 2, labels positive data on individual data of customers who have started using the product or service before the current time, and uses the product or service at the current time. Label negative data for individual customers who have not started That is, the individual data whose use start information is before the current time is labeled positive, and the individual data whose use start information is not before the current time is labeled negative. Then, the learning data generation unit 52 varies the number of positive individual data relative to the number of negative individual data in accordance with the degree of influence. In this example, the function value of the function of the influence degree is calculated, and the case where the individual data labeled positive is increased to the function value multiple is taken as an example.

  As a function of the degree of influence, for example, the ratio of the number of customers who have started using the product or service at the current time and the number of customers who have not started using the product or service at the current time You may use the function to do. That is, the functions exemplified below may be used.

  Impact x (Number of unused users at current time) / (Number of users at current time)

  In the above function, “the number of unused users at the current time” is the number of customers who have not yet started using the product or service at the current time (more specifically, the number of individual data). That is, in the customer data, the number of individual data whose usage start information is not before the current time. The “number of users at the current time” is the number of customers who have started using goods or services at the current time. That is, in the customer data, the number of individual data whose use start information is before the current time. In this way, a function may be used in which a ratio obtained by dividing “the number of unused users at the current time” by “the number of users at the current time” is an influence coefficient.

  Further, as a function of the influence degree, for example, a function having a coefficient of the influence degree as a ratio of the number of customers who start using the product or service at the current time and the total number of customers may be used. That is, the functions exemplified below may be used.

  Impact x (number of customers) / (number of users at the current time)

  In the above function, “number of customers” is the total number of customers (the total number of individual data in the customer data). The “number of unused users at the current time” is the same as in the case of the above function. In this way, a function may be used in which a ratio obtained by dividing “number of customers” by “number of users at the current time” is an influence coefficient.

  The learning data generation unit 52 increases the individual data labeled positive as the function value multiple of such a function. For example, suppose that j is a value obtained by substituting the input degree of influence into a function. In this case, a copy of the individual data labeled positive may be generated so that the number of individual data labeled positive is j times.

  Also, the value of the coefficient of influence in the two exemplified functions decreases as the number of users increases with time. The coefficient of influence is not limited to such a coefficient. Further, the influence value may be used as a function value as it is. In this case, the coefficient is 1.

  In the above example, the case where the individual data labeled positive is increased by a multiple of the function value, but in weighting, the number of individual data labeled positive is changed to the number of individual data labeled negative. What is necessary is just to change relatively with respect to it. This point is the same as the weighting performed by the previous time data group generation unit 32 in the first embodiment.

  After step B2, the classifier generation unit 53 generates a classifier using the learning data (step B3). The classifier generation unit 53 may generate a classifier by any one or a combination of data mining methods such as multiple regression analysis, decision tree, neural network, support vector machine, and Bayesian network.

  Further, as exemplified in the classifier generation operation of the previous time data group generation unit 32 in the first embodiment, a decision tree may be generated as a classifier. That is, the classifier generation unit 53 determines which item is first divided when learning data is given. At this time, the classifier generation unit 53 may calculate an evaluation value at the time of division for each of the items 1 to N and select an item having the largest evaluation value as an item most suitable for division. As the evaluation value, for example, the difference between the entropy of the node before division and the entropy after division may be used. The entropy of the node before the division and the entropy calculation method after the division are the same as the calculation method already described in the first embodiment, and the description is omitted. The classifier generation unit 53 performs the same process as described above for each node after the division, and sequentially repeats the process of determining which item to divide next, and when a predetermined condition is satisfied, Stop splitting. The classifier generation unit 53 generates a decision tree to be a classifier by pruning the tree-structured decision tree thus obtained. The predetermined conditions and the pruning process are the same as the predetermined conditions and the pruning process described in the first embodiment, and the description thereof is omitted. The decision tree (classifier) generated in this way increases the ratio of the individual data labeled with the first label in the learning data, and increases the ratio of the individual data labeled with the second label in the test data. Among them, there is a property that the frequency of determining to label the first label with respect to individual data having an item value similar to the item value of the individual data labeled with the first label is high. The classifier is not limited to the decision tree, and the classifier generation unit 53 may generate the classifier by an operation other than the above.

  After step B3, the test data label determination unit 54 predicts the label of the test data using the classifier generated in step B3 (step B4). That is, for each individual data in the test data, the test data label determination unit 54 collates the item indicating the customer attribute in the individual data with the classifier generated in step B3, and labels the individual data. Determine whether is positive or negative. For example, when the classifier is a decision tree, the test data label determination unit 54 refers to the item value of the individual data in the test data for the item of the node of the route, and traces the child node according to the item value. If the test data label determination unit 54 traces the leaf node, the test data label determination unit 54 may use the label having the larger count number in the leaf node as the determination result.

  The test data label determination unit 54 may display a positive or negative label determined for each individual data in the test data, for example, on a display device. Or you may make a printing apparatus print.

  According to the present embodiment, labeling is performed on a set of individual data including items indicating customer attributes, and a classifier is generated from learning data obtained by weighting. Then, the test data label determination unit 54 compares each item with the classifier for each individual data in the test data, and determines either positive or negative. That is, a classifier is generated using an item indicating the customer's attribute, and whether or not the customer starts using the item next to the current time from the item of the customer who has not started use and the classifier I'm predicting. Therefore, it is possible to determine whether or not the customer starts to use at the time after the current time by using the customer attribute.

  Further, since the label is predicted for each individual data, it is possible to predict whether or not to start using the product or service at the time next to the current time for each individual customer. In addition, if the current time to be designated is changed, it is possible to predict whether or not to start using a product or service for each customer at a desired time. That is, it is possible to predict whether or not an individual customer will start using a product or service at a certain time. Therefore, the prediction result can be used for sales promotion of goods and services.

  In addition, when predicting a diffusion curve by applying data to a logistic curve or the like, data from the rising edge (that is, data provided from the start of the provision of goods or services) is required. Even when the data is data in a partial period, it is possible to predict whether or not to start using the product or service for each customer at the time next to the current time.

  Next, a modification of the second embodiment is shown. In the following modification of the second embodiment, the prediction result for the test data is reflected in the customer data stored in the customer DB 2. Then, the current time is updated and the prediction for the test data is repeated to predict the degree of spread of the product or service.

  FIG. 14 is a block diagram illustrating a modification of the second embodiment. Constituent elements similar to those already described are denoted by the same reference numerals as those in FIG. In addition, FIG. 12 shows an example in which the spread prediction device 5 includes the input device 1 for inputting the current time, the degree of influence, and the like, and the output device 4 for outputting the prediction result. . The input device 1 is an input device such as a keyboard, and the output device 4 is an output device such as a display device.

  The spread prediction device 5 includes a test data generation unit 51, a learning data generation unit 52, a classifier generation unit 53, a test data label determination unit 54, and a current time update unit 55.

  In this modification, the test data label determination unit 54 overwrites and updates the use start information of individual data whose determination result in step B4 is different from the actual use start information. The case where the determination result is different from the actual use start information is a case where the use start information of the individual data indicates not used and the determination result is positive. In this case, the test data label determination unit 54 updates the use start information of the individual data in the customer data stored in the customer DB 2 to a time obtained by adding a certain amount of time (T) to the current time. . In this way, the time is updated so that the individual data indicates that the customer has started using the product or service from that time.

  The current time updating unit 55 updates the current time so that the time obtained by increasing the current time by a certain time (T) is set as the new current time after the test data label determination unit 54 determines the test data. To do. The current time update unit 55 is also realized by a CPU that operates according to, for example, a spread prediction program.

  The test data generation unit 51, the learning data generation unit 52, the classifier generation unit 53, and the test data label determination unit 54 update the customer data on condition that the updated current time is within a predetermined period. Then, the same processing as in the second embodiment is repeated using the current time.

  Next, the operation of this modification will be described. FIG. 15 is a flowchart illustrating an example of processing progress in a modification of the second embodiment. In this modified example, in addition to the current time and the degree of influence, a period to be targeted for diffusion prediction (hereinafter referred to as a prediction period) is input to the diffusion prediction apparatus 5. As the prediction period, for example, the last time of the period that is the target of the spread prediction may be input.

  Then, the process of step B1-B4 is performed. Steps B1 to B4 are the same as steps B1 to B4 (see FIG. 12) already described in the second embodiment.

  After step B4, the test data label determination unit 54 starts using the individual data in which the use start information of the individual data indicates that the individual data is not used, and a positive result is obtained as the determination result in step B4. Information is updated (step B5). The test data label determination unit 54 updates the use start information of the individual data in the customer data at a time obtained by adding a certain time T to the current time.

  Next, the current time update unit 55 updates the current time (step B6). In step B6, the current time update unit 55 sets a time obtained by adding a certain time T to the current time as a new current time.

  Then, the current time update unit 55 determines whether or not the updated current time is within the prediction period (step B7). If it is determined that the current time after the update is within the prediction period (Yes in Step B7), the spread prediction device 5 repeats the processes after Step B1.

  When it is determined that the updated current time exceeds the prediction period (No in step B7), the current time update unit 55 causes the output device 4 to output the prediction result (step B8). For example, the use disclosure information in the individual data for each customer may be output to the output device 4 as the predicted use start time of each customer. Further, the popularization curve of goods or services may be output to the output device 4. The spread prediction system may include a display device such as a display device as the output device 4, and the current time update unit 55 may display the prediction result on the output device 4. Alternatively, the spread prediction system may include a printing device as the output device 4, and the current time update unit 55 may cause the output device 4 to print the prediction result.

  When outputting a diffusion curve, for example, every time the process proceeds to step B1, the current time update unit 55 (which may be another component) counts the number of individual data of customers who have started using products or services at the current time May be counted from the customer DB, and the count value and the current time may be stored in association with each other. In step B8, for example, the current time is taken on the horizontal axis, the count value is taken on the vertical axis, a graph showing the transition of the count value accompanying the change in the current time is created, and the output device 4 outputs the graph. Good. This count value corresponds to the number of customers who have started use.

  In a modification of the second embodiment, a classifier is generated from learning data obtained by labeling a set of individual data including items indicating customer attributes and further weighting, and the classifier The prediction for the test data is repeated while updating the current time. Therefore, it is possible to predict the degree of the spread of products and services with high accuracy in consideration of differences in customer attributes (features).

  Moreover, the prediction accuracy of the spread can be improved as the customer data is increased.

  In addition, as a function used when generating learning data, the ratio between the number of customers who have started using the product or service at the current time and the number of customers who have not started using the product or service at the current time, or By using a function that takes the ratio of the number of customers who have started using products or services at the current time and the total number of customers as a coefficient of influence, depending on not only the influence but also the penetration rate of the product or service The function value can also be determined and weighted. As a result, prediction accuracy can be further increased.

  Further, as a function used at the time of generating learning data, a function including a coefficient corresponding to a customer attribute may be used as an influence coefficient. For example, weighting may be performed using the following function when generating learning data.

  Impact x Customer factor x (Number of unused users at current time) / (Number of users at current time)

  The learning data generation unit 52 may select a customer coefficient according to the content of the item indicating the customer attribute in the individual data, and use the customer coefficient to calculate a function value using, for example, the above function. . Thus, you may weight individually for every customer.

  Further, in the above-described modification example, the case of updating by overwriting the usage start information of the individual data stored in the customer DB 2 has been described as an example, but a copy of the usage start information included in the individual data from the beginning is described. May be copied as use start time prediction information, and the copy may be updated. In this case, the actual result and the predicted value can be held respectively.

Embodiment 3. FIG.
The spread prediction system according to the present embodiment calculates the degree of influence, and uses the degree of influence to determine whether or not each customer starts using the product or service at the prediction target time. FIG. 16 is a block diagram illustrating an example of a spread prediction system according to the present embodiment. The spread prediction system of this embodiment includes a customer DB 2, an influence degree estimation device 3, and a spread prediction device 5. Moreover, the input device 1 and the output device 4 may be provided. The influence degree estimation device 3 operates in the same manner as the influence degree estimation device 3 described in the first embodiment. Further, the spread prediction device 5 operates in the same manner as the spread prediction device 5 described in the second embodiment. The customer DB 2, the input device 1, and the output device 4 are the same as the customer DB 2, the input device 1, and the output device 4 described in the first embodiment and the second embodiment.

  The customer DB 2 is a storage means for storing customer data, and the customer data stored in the customer DB 2 is the same as the customer data in the first embodiment and the second embodiment. That is, customer data is a set of individual data for each customer including use start information and one or more items representing customer attributes other than use start information.

  The influence degree estimation device 3 receives the current time, the time interval, and a plurality of temporary influence degrees, for example, via the input device 1. The influence degree estimation device 3 estimates the influence degree using the current time, the time interval, and a plurality of temporary influence degrees, and passes the influence degree to the diffusion prediction device 5.

  The spread prediction device 5 receives the current time, for example, via the input device 1. The spread prediction device 5 performs prediction on the test data using the current time and the influence degree determined by the influence degree estimation device 3.

  In the present embodiment, the input current time is also used as an influence degree estimation target time in the influence degree estimation device 3.

  FIG. 17 is a block diagram showing in detail the spread prediction system of this embodiment. However, the input device 1 and the output device 4 are omitted.

  The influence level estimation device 3 includes a current time data generation unit 31, a previous time data group generation unit 32, a previous time classifier group generation unit 33, an error group calculation unit 34, and an influence level calculation unit 35. The spread prediction device 5 includes a test data generation unit 51, a learning data generation unit 52, a classifier generation unit 53, and a test data label determination unit 54.

  The current time data generation unit 31, the previous time data group generation unit 32, the previous time classifier group generation unit 33, the error group calculation unit 34, and the influence calculation unit 35 operate in the same manner as in the first embodiment. That is, steps A1 to A10 (see FIG. 3) are executed to determine the degree of influence. The influence degree calculation unit 35 passes the influence degree to the learning data generation unit 52.

  The test data generation unit 51, the learning data generation unit 52, the classifier generation unit 53, and the test data label determination unit 54 operate in the same manner as in the second embodiment. That is, Steps B1 to B4 (see FIG. 13) are executed, and it is determined for each customer whether or not the use of goods or services is started.

  In addition, the influence degree calculation part 35 determines one influence degree, and the spread prediction apparatus 5 should just perform the process after step B1 using the influence degree.

  Various modifications relating to the influence degree estimation device 3 described in the first embodiment may be applied to the influence degree estimation device 3 of the present embodiment. Moreover, the various deformation | transformation regarding the spread prediction apparatus 5 demonstrated in 2nd Embodiment may be applied to the spread prediction apparatus 5 of this embodiment. Similarly to the modification of the second embodiment (see FIG. 14), the spread prediction device 5 may include the current time update unit 55 and execute steps B1 to B8. For example, the popularization curve may be output to the output device 4.

  Current time data generation unit 31, previous time data group generation unit 32, previous time classifier group generation unit 33, error group calculation unit 34, influence calculation unit 35, test data generation unit 51, learning data generation unit 52, classifier The generation unit 53 and the test data label determination unit 54 are realized by, for example, a CPU that operates according to a spread prediction program. When the spread prediction system includes the current time update unit 55, the current time update unit 55 may also be realized by the CPU.

  Also in this embodiment, the same effect as that of the second embodiment can be obtained. In particular, since the influence estimation device 3 performs processing using the influence, the prediction accuracy can be further increased.

  In the second and third embodiments, the case where the spread prediction system includes the customer DB 2 has been described. However, the customer DB 2 is provided outside the spread prediction system, and the spread prediction system refers to the customer DB 2. Thus, a configuration may be used in which processing is performed.

Embodiment 4 FIG.
FIG. 18 is a block diagram illustrating an example of a spread prediction system according to the fourth embodiment of the present invention. The spread prediction system according to the fourth embodiment includes a customer DB 2, a threshold function estimation device 6, and a spread prediction device 7. Moreover, you may provide the input device and output device similar to the input device 1 and the output device 4 which were demonstrated in 1st Embodiment to 3rd Embodiment. The customer DB 2 is the same as the customer DB 2 described in the first to third embodiments.

  The customer DB 2 is a storage unit that stores customer data, and the customer data stored in the customer DB 2 is the same as the customer data in the first to third embodiments. That is, customer data is a set of individual data for each customer including use start information and one or more items representing customer attributes other than use start information.

  In the present embodiment, the threshold function estimation device 6 estimates a threshold function indicating the relationship between the penetration rate of goods or services and the threshold. The threshold function estimation device 6 obtains a penetration rate and a threshold at a plurality of times, and estimates a threshold function from the penetration rate and the threshold for each time. This threshold value is a threshold value for determining whether or not the first label is labeled on the individual data. The spread prediction device 7 generates test data from the customer data stored in the customer DB 2 and calculates a probability score that the first label is labeled on the individual data in the test data. The spread prediction device 7 also calculates the spread rate of the product or service, and calculates the threshold value by substituting the spread rate into the threshold function estimated by the threshold function estimation device 6. The spread prediction device 7 determines that the label for the individual data whose score is equal to or higher than the threshold is the first label, and the label for the individual data whose score is lower than the threshold is the second label. Hereinafter, the first label is described as positive (or +) and the second label is described as negative (or −), as in the embodiment described above.

  The threshold function estimation device 6 includes a reference time data group generation unit 61, a pre-threshold time data group generation unit 62, a pre-threshold time classifier group generation unit 63, an error group calculation unit 64, and a function estimation unit 65. With.

  The reference time data group generation unit 61 receives a plurality of reference times for determining threshold estimation target times, a time interval for designating a predetermined time before each reference time, and a plurality of temporary thresholds. . A plurality of reference times, time intervals, and a plurality of temporary thresholds may be input to the reference time data group generation unit 61 via an input device (not shown in FIG. 18) such as a keyboard.

  The reference time in the present embodiment is a reference time for determining a later-described threshold estimation time, which is a threshold estimation target time, and a threshold for each threshold estimation previous time is determined using later-described reference time data. The Note that a penetration rate is also calculated for each time before threshold estimation, and a threshold function is estimated from the relationship between the penetration rate and the threshold at each time before threshold estimation.

  The time interval is a time for designating a certain time before the input reference time. The input temporary threshold is a threshold candidate to be obtained by the threshold function estimation device 6. Each temporary threshold is a numerical value of 0 or more and 1 or less, for example.

  The number of reference times input to the reference time data group generation unit 61 is U, and the number of temporary thresholds is V.

  The reference time data group generation unit 61 generates one reference time data for one reference time. Therefore, U reference time data are generated. The reference time data is obtained by labeling the individual data of the customer who uses the product or service at the reference time as positive (first label), and the individual data of the customer who has not started the product or service data at the reference time. Data labeled negative (second label). The reference time data group generation unit 61 generates reference time data using the customer DB 2 (that is, based on customer data stored in the customer DB 2). The process of generating individual reference time data is the same as the process of generating current time data in the first embodiment. In addition, the reference time data group generation unit 61 obtains a time before the time interval from each reference time (a time that goes back by the time interval from the reference time). This time is referred to as a threshold pre-estimation time. The reference time data group generation unit 61 calculates the time before threshold estimation by subtracting the time interval from the reference time for each input reference time. Therefore, U times before threshold estimation are also obtained.

  The pre-threshold time data group generation unit 62 generates one pre-threshold time data for one pre-threshold time. Accordingly, U threshold value pre-estimation time data are generated. The pre-threshold time data group generation unit 62 generates pre-threshold time data for each pre-threshold time based on the pre-threshold time and the customer data stored in the customer DB 2. For the time before threshold estimation data, the individual database of customers who started using products or services at the time before threshold estimation is labeled positive, and individual customers who have not started using products or services at the time before threshold estimation The data is negative labeled. The pre-threshold time data is different from the previous time data described in the first embodiment in that no weighting is performed.

  Further, the threshold value pre-estimation time data group generation unit 62 calculates the penetration rate using the pre-threshold value time data. The penetration rate is the ratio of the number of customers who have started using goods or services by the time before threshold estimation with respect to the number of customers. The pre-threshold time data group generation unit 62 calculates the penetration rate for each pre-threshold time data. Therefore, U penetration rates are also calculated. The threshold value pre-estimation time data group generation unit 62 may store U penetration rates for each pre-threshold value time data for use by the function estimation unit 65.

  The pre-threshold time classifier group generation unit 63 generates one classifier based on one pre-threshold time data. Therefore, U classifiers are generated. The classifier in the present embodiment is a rule that determines a probability score that a first label is labeled on individual customer data.

  The error group calculation unit 64 calculates the score of each individual data in the reference time data using the classifier generated by the threshold value pre-estimation time classifier group generation unit 63, and compares the score with the temporary threshold value. Thus, the label of each individual data is predicted. This score is a probability score that the first label is labeled on the individual data. The error group calculation unit 64 performs this label prediction for each classifier. Further, in the label prediction with one classifier, label prediction is performed for each input V temporary thresholds. When performing the label prediction using a certain classifier and a temporary threshold, the error group calculation unit 64 calculates the score of each individual data in the reference time data with the classifier, and the score is equal to or higher than the temporary threshold. It is predicted that the label for the individual data is positive and the label for the individual data whose score is less than the provisional threshold is negative. Each individual data in the reference time data has already been actually labeled. The error group calculation unit 64 calculates an error between the label predicted using the classifier and the temporary threshold and the label actually labeled with the reference time data. The calculation of the error between the predicted label and the actual label is the same as the process of calculating the error between the predicted label and the actual label by the error group calculation unit 34 in the first embodiment. Since U classifiers are generated and errors are calculated for V temporary thresholds for each classifier, U × V errors are calculated.

  For each of the U classifiers, the function estimation unit 65 identifies the minimum error among the errors calculated for each of the V temporary thresholds, and corresponds the temporary threshold corresponding to the minimum error to the classifier. It is defined as a threshold value (that is, a threshold value corresponding to the time before threshold estimation). Specifically, when the number of provisional threshold values corresponding to the minimum error is one, the function estimation unit 65 determines the provisional threshold value as the threshold value. Further, when there are a plurality of temporary threshold values corresponding to the minimum error, the threshold value corresponding to the pre-threshold time is determined based on the plurality of temporary threshold values. For example, when there are a plurality of temporary threshold values corresponding to the minimum error, an average value of the plurality of temporary threshold values is calculated, and the average value is determined as a threshold value corresponding to the time before threshold estimation. Hereinafter, the case where there are a plurality of temporary threshold values corresponding to the smallest error and the average value of the temporary threshold values is set as the threshold value will be described as an example.

  Since a threshold is determined for each classifier, a threshold corresponding to each time before threshold estimation is determined. Further, the prevalence estimation time data group generation unit 62 calculates the penetration rate for each preestimation time. Thus, U sets of thresholds and penetration rates for each time before threshold estimation are obtained. The function estimation unit 65 estimates a threshold function from the set of the threshold and the penetration rate.

  The reference time data group generation unit 61, the pre-threshold time data group generation unit 62, the pre-threshold time classifier group generation unit 63, the error group calculation unit 64, and the function estimation unit 65 included in the threshold function estimation device 6 are, for example, It is realized by a CPU that operates according to a program. That is, the CPU that has read the program from the storage device provided in the threshold function estimation device 6 includes a reference time data group generation unit 61, a threshold value pre-estimation time data group generation unit 62, a threshold value pre-estimation time classifier group generation unit 63, The error group calculation unit 64 and the function estimation unit 65 may be operated.

  The spread prediction device 7 includes a test data generation unit 71, a labeling data generation unit 72, a classifier generation unit 73, and a test data label determination unit 74.

  The current time is input to the spread prediction device 7. This current time defines a determination target time as to whether or not the customer has started using the product or service, like the current time data described in the second embodiment. The current time may be input to the spread prediction device 7 via an input device (not shown in FIG. 18).

  The test data generation unit 71 generates test data including individual data of customers who have not started use at the current time based on the current time and the customer DB 2. Hereinafter, a case where the test data includes only individual data of customers who have not started using the product or service at the current time will be described as an example.

  The labeling data generation unit 72 generates labeling data based on the current time and customer data stored in the customer DB 2. For the labeling data, the individual data of the customer who has started using the product or service at the current time is labeled positive (first label), and the individual data of the customer who has not started using the product or service at the current time Is negative (second label). The labeling data is different from the learning data described in the second embodiment in that no weighting is performed. Except for whether or not weighting is performed, the labeling data generation processing in the present embodiment and the learning data generation processing in the second embodiment are the same.

  Further, the labeling data generation unit 72 calculates a penetration rate using the generated labeling data.

  The classifier generation unit 73 generates a classifier using the labeling data generated by the labeling data generation unit 72. This classifier is a rule for determining a probability score that the first label is labeled on the individual data of the customer, like the classifier generated by the pre-threshold time classifier group generation unit 63.

  The test data label determination unit 74 uses the classifier generated by the classifier generation unit 73 to calculate the score of each individual data in the test data. Further, the test data label determination unit 74 calculates a threshold value using the threshold function estimated by the function estimation unit 65 and the penetration rate calculated from the labeling data. That is, the threshold value is calculated by substituting the penetration rate into the threshold function. Then, the score is compared with the threshold value to determine the label of each individual data in the test data label. Specifically, the test data label determination unit 74 determines that the label for the individual data whose score is equal to or greater than the threshold among the individual data in the test data label is positive, and the individual whose score is less than the threshold It is determined that the label for the data is the second label. The customer of the individual data determined to be positive is predicted to start using the product or service at the time after the current time. The customer of the individual data determined to be negative is predicted not to start using the goods or services yet at the time after the current time. The time next to the current time is a time obtained by adding a predetermined fixed time to the current time.

  The test data label determination unit 74 may cause the output device (not shown in FIG. 18) to output a label prediction result (determination result) for each individual data in the test data.

  The test data generation unit 71, the labeling data generation unit 72, the classifier generation unit 73, and the test data label determination unit 74 included in the spread prediction device 7 are realized by a CPU that operates according to a spread prediction program, for example. That is, the CPU that has read the spread prediction program from the storage device provided in the spread prediction device 7 operates as the test data generation unit 71, the labeling data generation unit 72, the classifier generation unit 73, and the test data label determination unit 74. May be.

  Further, the threshold function estimation device 6 and the spread prediction device 7 may be realized by the same information processing device. Then, a reference time data group generation unit 61, a pre-threshold time data group generation unit 62, a pre-threshold time classifier group generation unit 63, an error group calculation unit 64, a function estimation unit 65, a test data generation unit 71, a labeling The data generation unit 72, the classifier generation unit 73, and the test data label determination unit 74 may be realized by a CPU that operates according to the popularization prediction program.

Next, the operation will be described.
FIG. 19 and FIG. 20 are flowcharts showing an example of processing progress of the threshold function estimation device 6 in this embodiment. For example, when a plurality of reference times, time intervals, and a plurality of temporary thresholds are input to the reference time data group generation unit 61 via an input device (not shown in FIG. 18), the spread prediction system is as follows. To work. A plurality of reference times are denoted as CTime [t]. The variable t is a variable for sequentially specifying a plurality of input reference times, and t = 1,. For example, if t = 1, it means that the first reference time is specified, and CTime [1] represents the first reference time. The time interval is denoted as Δt. A plurality of temporary thresholds are denoted as FalseThr [i]. Here, the variable i is a variable for sequentially specifying a plurality of input temporary thresholds, and i = 1,. For example, if i = 1, it means that the first temporary threshold value is specified, and FalseThr [1] represents the first temporary threshold value. The number of input reference times is U, and the number of temporary thresholds is V.

  The reference time data group generation unit 61 initializes a variable t that specifies the reference time so that t = 1. Since t = 1, the first reference time CTime [1] is selected. The reference time data group generation unit 61 subtracts the time interval Δt from CTime [1] to obtain the time before threshold estimation. That is, the pre-threshold time is set to CTime [1] −Δt (step C1). Next, the reference time data group generation unit 61 determines whether or not t ≦ U (step C2). If t ≦ U (Yes in Step C2), the process proceeds to Step C3. If t> U (No in Step C2), the process proceeds to Step C18.

  If it is determined in step C2 that t ≦ U, the reference time data group generation unit 61 generates reference time data based on the reference time CTime [t] determined by t and the customer data stored in the customer DB2. (Step C3). The reference time data group generation unit 61 reads customer data stored in the customer DB 2 and stores individual data of customers who have started using goods or services at the reference time CTime [t] (use start information is CTime [t] Generates reference time data by labeling positive data (individual data representing previous time) and negative data on individual data of customers who have not started using goods or services at the reference time CTime [t] To do. At this time, the reference time data group generation unit 61 uses, for example, the threshold pre-estimation time (CTime [t] −Δt) corresponding to CTime [t] among the individual data belonging to the customer data stored in the customer DB2. The reference time data is generated by excluding the individual data of the customer who has started using the product or service. That is, the individual data whose use start information is before the threshold estimation time is excluded from the customer data, and the remaining individual data is labeled positively or negatively to generate the reference time data. Hereinafter, a case will be described as an example in which the reference time data is generated by excluding the individual data of customers who have started to use at the time before threshold estimation. The reference time data generation process in step C3 is the same process as the current time data generation process in the first embodiment.

  Next to step C3, the threshold pre-estimation time data group 62 is based on the pre-threshold pre-estimation time (CTime [t] −Δt) determined by the variable t and the customer data stored in the customer DB 2. Data is generated (step C4). The threshold value pre-estimation time data group 62 reads customer data stored in the customer DB 2, labels positive data on individual data of customers who have started using goods or services before the pre-threshold value estimation time, In this case, the threshold data before threshold estimation is generated by labeling individual data of customers who have not started using the product or service with negative. That is, individual data whose use start information is before the time before threshold estimation is labeled positive, individual data whose use start information is not before the previous time is labeled negative, and time data before threshold estimation is Generate.

  Next, the threshold value pre-estimation time data group 62 calculates the penetration rate using the pre-threshold value time data (step C5). The penetration rate is the ratio of the number of customers who have started using goods or services by the time before threshold estimation with respect to the number of customers, and each individual data in the time data before threshold estimation corresponds to each customer. Accordingly, the ratio of the number of individual data labeled positive in the time data before threshold estimation to the total number of individual data in the time data before threshold estimation is the penetration rate. The threshold value pre-estimation time data group 62 may be obtained by dividing the number of individual data labeled positive in the pre-threshold time data by the total number of individual data in the pre-threshold time data. This penetration rate is denoted as D [t] using a variable t. The pre-threshold time data group 62 stores the calculated penetration rate D [t].

  After Step C5, the threshold value pre-estimation time classifier group generation unit 63 generates a classifier using the pre-threshold time data (Step C6). For example, the classifier generation unit 63 performs classification according to any one or a combination of data mining techniques whose outputs are continuous values such as multiple regression analysis, regression tree, neural network, support vector machine, Bayesian network, ensemble learning of decision tree, and the like. It is sufficient to generate a container.

  For example, the threshold pre-estimation time classifier group generation unit 63 may apply bagging, which is a type of ensemble learning, to a decision tree, generate a plurality of decision trees, and use the plurality of decision trees as a classifier. . An example of classifier creation processing in this case will be described. The pre-threshold estimation time classifier group generation unit 63 extracts some pieces of individual data from the pre-threshold estimation time data while allowing duplication, and generates a decision tree using the set of individual data. By repeating this process a plurality of times, a plurality of decision trees are generated, and the combination of the plurality of decision trees is set as one classifier. For example, assuming that the number of individual data extracted from the time data before threshold estimation is K, the time classifier group generation unit 63 before threshold estimation extracts K pieces of individual data by allowing duplication from the time data before threshold estimation. The decision tree may be generated repeatedly from the individual data. For example, if this iterative process is performed 500 times, 500 decision trees are created.

  In the following description, a case where a plurality of decision trees are generated by bagging as described above and the plurality of decision trees are used as a classifier is taken as an example.

  The process of generating a decision tree from the extracted individual data is the same as the process exemplified as the classifier generation operation of the previous time data group generation unit 32 in the first embodiment. That is, the threshold value pre-estimation time classifier group generation unit 63 determines which item (item representing the customer's attribute) is divided first. At this time, an evaluation value at the time of division may be calculated for each of items 1 to N, and an item having the largest evaluation value may be selected as an item most suitable for division. As the evaluation value, for example, the difference between the entropy of the node before division and the entropy after division may be used. The pre-threshold time classifier group generation unit 63 performs the same process as described above for each node after the division, and sequentially repeats the process of determining which item to divide next, so that a predetermined condition is satisfied. If it happens, the node division is stopped. The pre-threshold time classifier group generation unit 63 generates a decision tree by pruning the tree-structured decision tree obtained in this way. The predetermined conditions and the pruning process are the same as the predetermined conditions and the pruning process described in the first embodiment.

  In addition, the decision tree generated differs depending on the combination of K pieces of individual data extracted from the time data before threshold estimation. When applying a plurality of decision trees obtained as described above to individual data and obtaining the score (probability that the label will be positive) of the individual data, the label of the individual data is determined for each individual decision tree. A score is determined by dividing the number of times determined to be positive (used) by the number of times of determination. For example, when 500 decision trees are generated as a classifier, label determination is performed on individual data using each decision tree. As a result, if the number of times determined to be positive is X, X / 500 is obtained as a score.

  When the classifier is generated in step C6, the error group calculation unit 64 sets the initial value of the variable ErrorMin indicating the minimum value of the error, and the initial values i = 1 and p = 0 for the variables i and p. A value is set (step C7). The initial value of ErrorMin may be a maximum value that can be taken by the threshold calculation unit 64 or a value that is sufficiently larger than a value that can be taken by the error. The variable p is a variable for designating a temporary threshold corresponding to the minimum error. For example, if there are p provisional threshold values corresponding to the error of the minimum word, the first to pth provisional threshold values corresponding to the minimum error are respectively represented as Thres [1], ..., Thres [p]. To do.

  After step C7, the error group calculation unit 64 determines whether the temporary threshold number i is equal to or smaller than the temporary threshold number V (step C8). If i ≦ V (Yes in Step C8), the process proceeds to Step C9. If i> V (No in Step C8), the process proceeds to Step C16.

  In step C9, the error group calculation unit 64 uses the pre-threshold time classifier generated in step C6 and the temporary threshold FalseThr [i], and each individual data in the reference time data generated in step C3. Is predicted (step C9). In this example, the error group calculation unit 64 compares, for each individual data in the reference time data, an item indicating the customer attribute in the individual data and each decision tree generated as a classifier, Determine labels for individual data. Since there are a plurality of decision trees, the error group calculation unit 64 performs this determination for each decision tree, and obtains the ratio of the number of times determined to be positive with respect to the number of determinations (number of determined trees) as the score of the individual data. That is, the error group calculation unit 64 calculates “the number of times determined to be positive / the number of determinations” for one piece of individual data, and uses the value as a score. Further, the error group calculating unit 64 compares the temporary threshold FalseThr [i] determined by the variable i with the score, and predicts that the label for the individual data is positive if the score is equal to or greater than FalseThr [i]. If the score is less than FalseThr [i], it is predicted that the label for the individual data is negative.

  Next, the error group calculation unit 64 calculates an error between the prediction result of the label labeled on each individual data in the reference time data and the reference time data (step C10). That is, since each individual data in the reference time data is labeled with a positive or negative label when the reference time data is generated, it is labeled with the prediction result in step C9 and the actual reference time data. Predict the error from the label. This error is referred to as Error.

  In step C10, the error group calculation unit 64 compares, for example, a label predicted for each individual data in the reference time data with a label actually labeled on each individual data in the reference time data. Different numbers of individual data may be set as Error.

  Alternatively, the error group calculation unit 64 actually labels positive in the number of individual data predicted to be labeled positive in step C9 among the individual data in the reference time data and the reference time data generated in step C3. The absolute value of the difference from the number of attached individual data may be Error.

  After step C10, the function estimation unit 65 compares the error Error calculated in step C10 with the minimum error value ErrorMin, and determines whether Error is less than ErrorMin (step C11). When Error is less than ErrorMin (Yes in Step C11), the function estimation unit 65 initializes p to 0 and substitutes Error for ErrorMin (Step C13). In step C13, the function estimation unit 65 initializes the variable Sum to 0. The variable Sum is a variable for storing the sum of provisional threshold values that minimize the error, and is used to calculate the average value of the provisional threshold values. If Error is less than ErrorMin, it means that a minimum value smaller than the error that has been minimized so far has been found. In this case, in Step C13, ErrorMin is updated with the minimum value, and when there are a plurality of temporary threshold values corresponding to the error, a variable p for individually specifying each temporary threshold value is initialized.

  After step C13, the function estimation unit 65 updates Sum by incrementing the value of p by 1 and adding the provisional threshold FalseThr [i] to the value of Sum (step C14). That is, the Sum is updated so that the value of Sum + FalseThr [i] becomes the new Sum value. When it is determined that Error is equal to or greater than ErrorMin (No in step C11), the function estimation unit 65 determines whether Error is equal to ErrorMin (step C12). Even when it is determined that they are equal (Yes in Step C12), the function estimation unit 65 increments the value of p by 1, adds FalseThr [i] to the Sum value, and updates the Sum (Step C14). .

  When the process proceeds from step C13 to step C14, p = 1, and the temporary threshold value FalseThr [i] currently focused on is set as the first temporary threshold value that minimizes the error. In Step C12, Error = ErrorMin, and when the process proceeds to Step C14, p becomes a value of 2 or more. In this case, one or more provisional threshold values corresponding to the minimum error have already been found, and the function estimation unit 65 sets the currently-thought provisional threshold value FlameThr [i] as the p-th provisional threshold value that minimizes the error. It will be determined.

  After step C14 or when it is determined in step C12 that Error is not equal to ErrorMin, the function estimation unit 65 increments a variable i indicating a temporary threshold (step C15). Thereafter, the process of step C8 is repeated using i incremented in step C15.

  If it is determined in step C8 that i> V, the function estimation unit 65 sets the value of the threshold corresponding to t (denoted as Thres [t]) to Sum / p (step C16). That is, Sum, which is the sum of the temporary threshold values corresponding to the minimum error, is divided by the number p of the temporary threshold values, and the average value is set as the threshold value. If there is only one temporary threshold corresponding to the minimum error, Sum is the value of the temporary threshold, and the temporary threshold is set to Thres [t].

  After step C16, the function estimating unit 65 increments a variable t for sequentially specifying the reference time (step C17). After step C17, the process after step C2 is repeated using the incremented t.

  When it is determined in step C2 that t> U, the function estimation unit 65 determines a threshold value that is a function of a threshold value for the penetration rate from the penetration rate D [t] and the threshold value Thres [t] obtained for each time before the threshold estimation. A function is obtained (step C18). At the time of shifting to Step C18, the time between threshold estimations at each t is calculated, and the penetration rate D [t] and the threshold Thres [t] corresponding to the time before the threshold estimation are obtained. That is, a set of the penetration rate D [t] and the threshold value Thres [t] is obtained. The function estimation unit 65 may obtain a threshold function by approximating the relationship between the penetration rate D [t] and the threshold value Thres [t] by a curve. For example, assuming that Thres [t] is expressed by the following equation (1) using D [t], the parameters a and b of the equation (1) may be estimated.

  Thres [t] = a · D [t] + b Formula (1)

  In equation (1), a and b are parameters to be estimated. U combinations of the penetration rate D [t] and the threshold value Thres [t] are obtained, and the function estimating unit 65 uses, for example, the U combinations (a combination of the threshold value and the penetration rate) to calculate the least square method. The parameters a and b may be calculated by Note that the value of the parameter a is low if the penetration rate of goods or services increases rapidly with time, and increases if the increase in penetration rate is moderate. For example, a is a value such as 0.5. The parameter b is a value that is low if the diffusion rate at the diffusion start stage suddenly increases, and is low if the diffusion rate hardly increases, and is a value from 0 to 1 and a small value such as 0.05.

  Moreover, Formula (1) is an example of a threshold function whose parameters are not determined, and the threshold function including the parameter to be estimated is not limited to Formula (1). The threshold function including the parameter to be estimated exemplified in Expression (1) may be input to the user of the spread prediction system. Alternatively, it may be determined in advance.

  The function estimation unit 65 determines the threshold function by calculating the parameters a and b by, for example, the least square method. The function estimation unit 65 inputs the threshold function to the spread prediction device 7 (test data label determination unit 74).

  FIG. 21 is a flowchart illustrating an example of processing progress of the spread prediction device 7 in the present embodiment. For example, when the threshold value function is input from the function estimation unit 65 and the current time is input via an input device such as a keyboard, the spread prediction device 7 operates as follows.

  The test data generation unit 71 generates test data (step D1). This operation is the same as step B1 in the second embodiment. Here, an example will be described in which test data including only individual data of customers who have not started using products or services at the current time is generated. The test data generation unit 71 selects only individual data of customers who have not started using the product or service at the current time from the customer data stored in the customer DB 2 (that is, the time when the use start information is before the current time). Non-individual data only) and the set of the individual data is used as test data.

  Next, the labeling data generation unit 72 generates labeling data (step D2). The labeling data generation unit 72 reads the customer data stored in the customer DB 2, labels the individual data of the customer who has started using the product or service before the current time, and labels the product or service at the current time. Labeling data is generated by labeling individual customer data that has not been used negatively.

  Next, the labeling data generation unit 72 calculates a penetration rate from the labeling data (step D3). Since each individual data in the labeled data corresponds to each customer, the penetration rate is the ratio of the number of individual data labeled positive in the labeled data to the total individual data in the labeled data . The labeling data generation unit 72 may obtain the penetration rate by dividing the number of individual data labeled positive in the labeling data by the total number of individual data in the labeling data.

  Next, the classifier generation unit 73 generates a classifier using the labeling data (step D4). This classifier is the same as the classifier generated by the threshold value pre-estimation time classifier group generation unit 63 of the threshold function estimation device 6. In this example, a plurality of decision trees are generated by bagging, and the plurality of decision trees are assumed to be classifiers. That is, the classifier generation unit 73 generates a plurality of decision trees by extracting a part of individual data while allowing duplication from the labeled data and repeating the process of generating a decision tree using the set of the individual data. To do. Assuming that the number of individual data extracted from the labeling data is K, the classifier generation unit 73 extracts K pieces of individual data allowing duplication from the labeling data, and generates a decision tree from the K pieces of individual data. Can be repeated.

  The process of generating a decision tree from the extracted individual data is the same as the process exemplified as the classifier generation operation of the previous time data group generation unit 32 in the first embodiment. The classifier generation unit 73 determines which item (item representing the customer's attribute) is to be divided first. At this time, an evaluation value at the time of division may be calculated for each of items 1 to N, and an item having the largest evaluation value may be selected as an item most suitable for division. As the evaluation value, for example, the difference between the entropy of the node before division and the entropy after division may be used. The classifier generation unit 73 performs the same process as described above for each node after the division, and sequentially repeats the process for determining which item is to be divided next, and when a predetermined condition is satisfied, Stop splitting. The classifier generation unit 73 generates a decision tree by pruning the tree-structured decision tree thus obtained. The predetermined conditions and the pruning process are the same as the predetermined conditions and the pruning process described in the first embodiment.

  The test data label determination unit 74 substitutes the penetration rate calculated in step D3 for the threshold function estimated by the function estimation unit 65 to calculate the threshold (step D5).

  Next, the test data label determination unit 74 predicts the label of the test data using the classifier generated in step D4 and the threshold value calculated in step D5 (step D6). In step D6, the test data label determination unit 74 performs the following processing for each individual data in the test data. The test data label determination unit 74 compares an item indicating the customer attribute in the individual data with each decision tree generated as a classifier, and determines a label for the individual data. Since there are a plurality of decision trees, the test data label determination unit 74 performs this determination for each decision tree, and obtains the ratio of the number of times determined to be positive with respect to the number of determinations (number of determined trees) as the score of the individual data. That is, the test data label determination unit 74 calculates “the number of times determined to be positive / the number of determinations” for one piece of individual data, and uses the value as a score. Furthermore, the test data label determination unit 74 compares the threshold calculated in step D5 with the score, and if the score is equal to or greater than the threshold, determines that the label for the individual data is positive, and the score is less than the threshold. If so, it is determined that the label for the individual data is negative.

  The test data label determination unit 74 may display the positive or negative label determined for each individual data in the test data on a display device or the like, or may cause the printing device to print.

  According to the present embodiment, similarly to the second embodiment, it is possible to predict for each customer when the customer starts using the product or service. In the present embodiment, when obtaining a threshold function, a prediction error (Error) is obtained for each of a plurality of input reference times and for each of a plurality of temporary thresholds, and a threshold value that minimizes the error is obtained. Since the threshold function is estimated, the acceptability that changes every moment can be estimated as the threshold function. In addition, when performing weighting, a classifier is generated every time weighting is performed. However, in this embodiment, since weighting is not performed, when calculating an error (Error) for each provisional threshold, There is no need to generate a classifier for each temporary threshold, and an error in each temporary threshold can be obtained using one classifier. Therefore, the calculation time for generating the classifier can be shortened.

  Next, a modification of the fourth embodiment is shown. In the following modification of the fourth embodiment, the current time is updated and prediction for test data is repeated, as in the modification of the second embodiment (see FIGS. 14 and 15).

  FIG. 22 is a block diagram illustrating a modification of the fourth embodiment. The same components as those already described are denoted by the same reference numerals as those in FIG. 18, and detailed description thereof is omitted. The spread prediction device 7 includes a test data generation unit 71, a labeling data generation unit 72, a classifier generation unit 73, a test data label determination unit 74, and a current time update unit 75. The current time update unit 75 is the same as the current time update unit 55 shown in the modification of the second embodiment.

  In this modification, the test data label determination unit 74 overwrites and updates the use start information of the individual data whose determination result in step D6 is different from the actual use start information. That is, if the use start information of the individual data indicates that it is not used and it is determined to be positive in Step D6, the test data label determination unit 74 starts using the individual data in the customer data stored in the customer DB2. Information is updated to a time obtained by adding a certain time (T) to the current time.

  The current time updating unit 75 updates the current time so that the time obtained by increasing the current time by a certain time (T) is set as the new current time after the test data label determination unit 74 determines the test data. To do. The current time update unit 75 is also realized by, for example, a CPU that operates according to a popularization prediction program.

  When the current time is updated, the test data generation unit 71, the labeling data generation unit 72, the classifier generation unit 73, and the test data label determination unit 74 use the updated current time and the updated customer data, The processing after step D1 is repeated.

  Note that the condition for ending this iterative process may be, for example, that the updated current time exceeds a predetermined period (predicted period), as in the modification of the second embodiment. The current time update unit 75 may update the current time on condition that the updated current time is within the prediction period. The prediction period may be input in advance by the user of the popularization prediction system, for example.

  If the current time update unit 75 determines that the updated current time exceeds the prediction period, the current time update unit 75 may cause the output device (not shown in FIG. 22) to output the prediction result. This output mode is the same as the modification of the second embodiment illustrated in FIG. For example, a popularization curve may be output. In this case, every time the process proceeds to step D1, the current time update unit 75 (which may be another component) calculates the number of individual data of the customer who has started using the product or service at the current time from the customer DB. It is only necessary to count and store the count value and the current time in association with each other. Then, the current time updating unit 75 may create and output a graph showing the transition of the count value accompanying the change in the current time, with the current time on the horizontal axis and the count value on the vertical axis.

  In this modification, instead of overwriting the use start information of the individual data stored in the customer DB 2, a copy of the use start information included in the individual data from the beginning is copied as use start time prediction information. The copy may be updated.

  FIG. 23 is a block diagram illustrating another modification of the fourth embodiment. 18 and 22, the case where the spread prediction system includes the threshold function estimation device 6 has been described. However, as illustrated in FIG. 23, a configuration that does not include the threshold function estimation device 6 may be used. The operations of the test data generation unit 71, labeling data generation unit 72, classifier generation unit 73, and test data label determination unit 74 shown in FIG. 23 are the same as those already described. In a configuration that does not include the threshold function estimation device 6, a threshold function may be input to the test data label determination unit 74 from the outside of the spread prediction system. For example, a user of the spread prediction system may input a threshold function. Also in this modification, the threshold function estimation device 6 may include a current time update unit 75.

  In this embodiment, gender, age, purchase history of existing products, innovator characteristics, and contact media are items indicating customer characteristics, and customer data including use start time is stored in the customer DB 2 in advance together with these items. . With regard to individual data of customers who have started using the product during the four years from the release date of the specific product, the customer has set the use start time of the product as use start information. For individual data of customers who have not yet started using the product during the four years, “?” Meaning “not used” is set as the usage start information. The customers whose purchase time (use start time) is described as use start information is 8.4% of the total number of customers stored in the customer DB 2, and “?” Meaning “unused” is described as use start information. The customers were 91.6% of the total number of customers stored in the customer DB2. In addition, for each customer, the above items indicating customer characteristics were defined. When the value of an item is unknown, "?" Meaning unknown is described as the value of the item.

  The customer data as described above was previously stored in the customer DB 2 and applied to the influence degree estimation system of the first embodiment. As a result, as a result of the influence degree estimation system estimating the influence degree, the influence degree was 0.191.

  Moreover, this influence degree 0.191 was applied to the spread prediction system (FIG. 14) of 2nd Embodiment. An example of the diffusion curve output by the diffusion prediction system is shown in FIG. The penetration rate in the 4th year after the start of sales is 8.4%, and the function value at this time is 0.191 × (100−8.4) /8.4≈2, and the current time in the 4th year According to the prediction, the customers who have started to use are weighted twice as much as the unused customers. Even when the subsequent time is the current time, the influence level is constant at 0.191, but the function value decreases as the penetration rate increases and the weighting is also reduced.

  In addition, in order to verify the penetration curve obtained by the penetration prediction system, we also investigated the penetration results for two years from the 4th year to the 6th year after product launch.

  The solid line shown in FIG. 24 is a diffusion curve output by the diffusion prediction system. In addition, the curve plotted with + indicates the results of popularization up to the fourth year after product release. The curve plotted with a rectangle shows the results of popularization for 2 years from the 4th year to the 6th year after product launch. As shown in FIG. 24, the diffusion curve output by the diffusion prediction system (solid line shown in FIG. 24) was very close to the actual achievement of diffusion (rectangular plot shown in FIG. 24).

  The present invention is preferably applied to, for example, a spread prediction system that predicts the use start time of goods and services for each individual customer. Further, the present invention is suitably applied to an influence degree estimation system that estimates an influence degree at a designated time.

It is a block diagram which shows the example of the influence degree estimation system of this invention. It is a block diagram which shows the example of the influence estimation system provided with the input device and the output device. It is a flowchart which shows the example of the process progress of the influence degree estimation system of this invention. It is explanatory drawing which shows the example of customer data. It is explanatory drawing which shows the example of present time data. It is explanatory drawing which shows the example of the data before weighting is performed in the previous time data generation process. It is explanatory drawing which shows the example of previous time data. It is explanatory drawing which shows the example of the combination of a known independent variable and a dependent variable. It is explanatory drawing which shows the example of a classifier. It is explanatory drawing which shows the example of the dependent variable estimated from a classifier. It is explanatory drawing which shows the error in each temporary influence degree calculated | required for every present time. It is a block diagram which shows the example of the spread prediction system of this invention. It is a flowchart which shows the example of the process progress of the spread prediction system of this invention. It is a block diagram which shows the modification of 2nd Embodiment. It is a flowchart which shows the example of the process progress in the modification of 2nd Embodiment. It is a block diagram which shows the example of the spread prediction system of 3rd Embodiment. It is a block diagram which shows the penetration prediction system of 3rd Embodiment in detail. It is a block diagram which shows the example of the spread prediction system of 4th Embodiment. It is a flowchart which shows the example of a process progress of the threshold value function estimation apparatus in 4th Embodiment. It is a flowchart which shows the example of a process progress of the threshold value function estimation apparatus in 4th Embodiment. It is a flowchart which shows the example of a process progress of the spread prediction apparatus in 4th Embodiment. It is a block diagram which shows the modification of 4th Embodiment. It is a block diagram which shows the other modification of 4th Embodiment. It is a graph which shows the example of the penetration curve which the penetration prediction system outputted.

Explanation of symbols

2 customer database 31 current time data generation unit 32 previous time data group generation unit 33 previous time classifier group generation unit 34 error group calculation unit 35 influence calculation unit 51 test data generation unit 52 learning data generation unit 53 classifier generation unit 54 Test data label determination unit 55 Current time update unit 61 Reference time data group generation unit 62 Time data group generation unit before threshold estimation 63 Time classifier group generation unit before threshold estimation 64 Error group calculation unit 65 Function estimation unit 71 Test data generation unit 72 Labeling Data Generation Unit 73 Classifier Generation Unit 74 Test Data Label Determination Unit

Claims (36)

When the customer starts using the product or service, the use start information indicates the use start time of the product or service, and when the product or service is not used, the use start information indicating that the product or service is not used, and the use start A customer database that stores customer data that is a collection of individual data for each customer including one or more items representing customer attributes other than information;
Using the customer database, a test data generating unit that generates test data that is data including individual data of customers who have not started using goods or services at the current time for determining the determination target time;
Label individual data of customers who have started using the goods or services at the current time, label second data to individual data of customers who have not started using the goods or services, This is data in which the number of individual data labeled with the first label is changed according to the degree of influence, which is the degree that the customer of the product or service urges others to use the product or service within a certain period. A learning data generation unit for generating learning data;
A classifier generator that generates a classifier, which is a rule for determining which of the customer's individual data is labeled with the first label or the second label from the item representing the customer's attribute, based on the learning data; ,
A spread prediction system comprising: a test data label determination unit that determines a label for each individual data in the test data from the classifier and each individual data item in the test data. A current time update unit that sets a new current time as a time obtained by increasing the current time by a certain amount of time;
The test data label determination unit gives information indicating the use start time to the individual data determined to be labeled with the first label,
When the current time is updated, the test data generation unit generates test data at the current time after the update,
When the current time is updated, the learning data generation unit generates learning data at the current time after the update,
When the current time is updated and new learning data is generated, the classifier generator generates a new classifier based on the learning data,
When the current time is updated and new test data is generated, the test data label determination unit determines whether the individual data in the test data from the classifier and the individual data items in the test data. The spread prediction system according to claim 1, wherein a label is determined. The test data generation unit generates test data from customer data stored in the customer database, excluding individual data of customers who have started using goods or services at the current time. The spread prediction system described. The learning data generation unit is an individual data in which the first label is labeled as a function value multiple of a function having a ratio of the number of customers who have started using a product or service and the number of customers who have not started yet as a coefficient of influence. The spread prediction system according to any one of claims 1 to 3, wherein learning data is generated by increasing the number. The learning data generation unit increases the number of individual data labeled with the first label to a function value multiple of the function having the ratio of the number of customers who have started using the product or service and the total number of customers as a coefficient of influence. The diffusion prediction system according to any one of claims 1 to 3, wherein learning data is generated. The classifier generation unit labels the first label among the individual data labeled with the second label in the test data as the ratio of the individual data labeled with the first label in the learning data increases. 6. A classifier that generates a high frequency of determining that the first label is labeled with respect to individual data having an item value similar to the item value of the attached individual data is generated. The spread prediction system according to claim 1. The diffusion prediction system according to any one of claims 1 to 6, wherein the learning data generation unit generates learning data using an influence degree input from outside the diffusion prediction system. The current time that is used as the target time for determining the determination target time and the degree of influence estimation, the time interval for designating a predetermined time before the current time, and a plurality of temporary influence degrees that are candidates for the influence degree are input. Calculating a previous time that is a time before the time interval from the current time, and using a customer database, a first label is added to individual data of a customer who has started using the product or service at the current time A current time data generation unit for generating current time data by labeling and labeling a second label on individual data of a customer who has not started using the product or service;
For each temporary influence degree, the first label is labeled on the individual data of the customer who has started using the product or service at the previous time, and the individual data of the customer who has not started using the product or service A previous time data group generation unit that generates the previous time data, which is data obtained by labeling the second label and changing the number of individual data labeled with the first label according to the temporary influence degree;
Based on the previous time, a classifier that is a rule for determining whether to label individual customer data from the item representing the customer attribute for each individual previous time data. A previous time classifier group generation unit to generate;
For each classifier generated for each previous time data, predict the label that is labeled on the individual data in the current time data from the classifier and the individual data items in the current time data, and predict An error group calculation unit for calculating an error between the result and the current time data;
Among the errors calculated for each classifier, the smallest error is specified, and when there is one temporary influence corresponding to the smallest error, the temporary influence is determined as the influence, 7. The apparatus according to claim 1, further comprising an influence degree calculation unit that determines an influence degree based on the plurality of temporary influence degrees when there are a plurality of temporary influence degrees corresponding to the errors. The spread prediction system described in 1. The current time data generation unit generates current time data from a set of individual data excluding the individual data of the customer who started using the product or service at the previous time among the customer data stored in the customer database. The spread prediction system according to claim 8. The previous time data group generator is a function value multiple of the function that uses the ratio of the number of customers who have started using goods or services at the previous time and the number of customers who have not started at the previous time as a coefficient of temporary impact. The spread prediction system according to claim 8 or 9, wherein the previous time data is generated by increasing the number of individual data labeled with the first label. The previous time data group generation unit labels the first label as a function value multiple of a function having the ratio of the number of customers who have started using the product or service at the previous time and the total number of customers as a coefficient of temporary impact. The spread prediction system according to claim 8 or 9, wherein the previous time data is generated by increasing the number of attached individual data. For each classifier generated for each previous time data, the error group calculation unit labels the individual data in the current time data from the classifier and the individual data items in the current time data. 12. The label is predicted, and the number of individual data in which the predicted label and the label actually labeled with the current time data are different is calculated as an error. The spread prediction system described in 1. For each classifier generated for each previous time data, the error group calculation unit labels the individual data in the current time data from the classifier and the individual data items in the current time data. 9. The label is predicted, and the absolute value of the difference between the number of individual data predicted that the first label is labeled and the number of individual data labeled with the first label in the current time data is calculated as an error. The spread prediction system according to any one of claims 11 to 11. When the customer starts using the product or service, the use start information indicates the use start time of the product or service, and when the product or service is not used, the use start information indicating that the product or service is not used, and the use start A customer database that stores customer data that is a collection of individual data for each customer including one or more items representing customer attributes other than information;
Specify the current time used as an estimation target time of the degree of influence, which is the degree to which the customer of the product or service urges others to use the product or service within a certain period, and a certain time before the current time Time interval and a plurality of temporary influence levels that are candidates for influence level are input, a previous time that is a time before the time interval is calculated from the current time, and the current time is calculated using the customer database. Current data is obtained by labeling individual data of customers who have started using the product or service with a first label and labeling individual data of customers who have not started using the product or service with a second label. A current time data generation unit for generating
For each temporary influence degree, the first label is labeled on the individual data of the customer who has started using the product or service at the previous time, and the individual data of the customer who has not started using the product or service A previous time data group generation unit that generates the previous time data, which is data obtained by labeling the second label and changing the number of individual data labeled with the first label according to the temporary influence degree;
Based on the previous time, a classifier that is a rule for determining whether to label individual customer data from the item representing the customer attribute for each individual previous time data. A previous time classifier group generation unit to generate;
For each classifier generated for each previous time data, predict the label that is labeled on the individual data in the current time data from the classifier and the individual data items in the current time data, and predict An error group calculation unit for calculating an error between the result and the current time data;
Among the errors calculated for each classifier, the smallest error is specified, and when there is one temporary influence corresponding to the smallest error, the temporary influence is determined as the influence, An influence degree estimation system comprising: an influence degree calculation unit that determines an influence degree based on the plurality of provisional influence degrees when there are a plurality of provisional influence degrees corresponding to errors. The current time data generation unit generates current time data from a set of individual data excluding the individual data of the customer who started using the product or service at the previous time among the customer data stored in the customer database. The influence estimation system according to claim 14. The previous time data group generator is a function value multiple of the function that uses the ratio of the number of customers who have started using goods or services at the previous time and the number of customers who have not started at the previous time as a coefficient of temporary impact. The influence estimation system according to claim 14 or 15, wherein the previous time data is generated by increasing the number of individual data labeled with the first label. The previous time data group generation unit labels the first label as a function value multiple of a function having the ratio of the number of customers who have started using the product or service at the previous time and the total number of customers as a coefficient of temporary impact. The influence estimation system according to claim 14 or 15, wherein the previous time data is generated by increasing the number of attached individual data. For each classifier generated for each previous time data, the error group calculation unit labels the individual data in the current time data from the classifier and the individual data items in the current time data. 18. The label is predicted, and the number of individual data in which the predicted label and the label actually labeled with the current time data are different is calculated as an error. The impact estimation system described in. For each classifier generated for each previous time data, the error group calculation unit labels the individual data in the current time data from the classifier and the individual data items in the current time data. The label is predicted, and the absolute value of the difference between the number of individual data predicted to be labeled with the first label and the number of individual data with the first label labeled with the current time data is calculated as an error. The influence degree estimation system according to claim 1. When the customer starts using the product or service, the use start information indicates the use start time of the product or service, and when the product or service is not used, the use start information indicating that the product or service is not used, and the use start A customer database that stores customer data that is a collection of individual data for each customer including one or more items representing customer attributes other than information;
Using the customer database, a test data generating unit that generates test data that is data including individual data of customers who have not started using goods or services at the current time for determining the determination target time;
The first label is labeled on individual data of customers who have started using the product or service at the current time, and the second label is labeled on individual data of customers who have not started using the product or service. A labeling data generating unit that generates labeling data and calculates a ratio of the number of individual data labeled with the first label in the labeling data as a penetration rate of the product or service;
A classifier generating unit that generates a classifier that is a rule for determining a probability score that the first label is labeled on the individual data of the customer from the item representing the attribute of the customer based on the labeling data;
By substituting the penetration rate calculated by the labeling data generation unit into a threshold function that is a function of the threshold value of the score for the penetration rate, the threshold value is calculated, and from the individual data items in the classifier and the test data, A score of the probability that the first label is labeled to each individual data in the test data is determined, and it is determined that the label for the individual data in which the score is equal to or higher than the threshold among the individual data in the test data is the first label And a test data label determination unit that determines that the label for the individual data having a score less than the threshold is the second label. A current time update unit that sets a new current time as a time obtained by increasing the current time by a certain amount of time;
The test data label determination unit gives information indicating the use start time to the individual data determined to be labeled with the first label,
When the current time is updated, the test data generation unit generates test data at the current time after the update,
When the current time is updated, the labeling data generation unit generates the labeling data at the current time after the update, calculates the diffusion rate in the labeling data,
When the current time is updated and new labeling data is generated, the classifier generator generates a new classifier based on the labeling data,
When the current time is updated and new test data is generated, the test data label determination unit converts each individual data in the test data from the classifier and each individual data item in the test data. A score of the probability that the first label is labeled is determined, a threshold value is newly calculated by substituting the penetration rate calculated by the labeling data generation unit into the threshold function, and the score among the individual data in the test data is The spread prediction system according to claim 20, wherein a label for individual data that is equal to or greater than a threshold is determined to be a first label, and a label for individual data having a score less than the threshold is determined to be a second label. The test data generation unit generates test data excluding individual data of customers who have started using goods or services at the current time from customer data stored in a customer database. The spread prediction system described. A plurality of reference times for determining a threshold estimation target time, a time interval for designating a predetermined time before each reference time, and a plurality of temporary thresholds that are threshold candidates are input, and for each reference time , Calculate the time before threshold estimation that is the time before the time interval from the reference time, and use the customer database to put the first label on the individual data of the customer who has started using the goods or services at the reference time A reference time data group generation unit for generating reference time data for labeling and labeling a second label on individual data of a customer who has not started using the product or service;
For each individual pre-threshold estimation time, a first label is labeled on the individual data of the customer who has started using the product or service at the pre-threshold estimation time. Generate time data before threshold estimation, which is data obtained by labeling individual data with a second label, and determine the ratio of the number of individual data labeled with the first label in the time data before threshold estimation. Time threshold value pre-estimation data group generation unit to calculate as a rate,
For each pre-threshold time estimation data, a classifier that is a rule for determining the probability score that the first label is labeled on the individual data of the customer from the item representing the customer attribute is used as the pre-threshold time data. A threshold value pre-estimation time classifier group generation unit to be generated based on:
For each classifier generated for each time data before threshold estimation, a first label is labeled on each individual data in the reference time data from the classifier and each individual data item in the reference time data. The probability score to be attached is determined, and for each temporary threshold, the label of the individual data whose score is equal to or higher than the temporary threshold is predicted to be the first label, and the label of the individual data whose score is less than the threshold is the second label An error group calculation unit that predicts that there is an error and calculates an error between the prediction result and the reference time data;
The minimum error among the errors calculated for each temporary threshold is specified for each individual classifier, and when there is one temporary threshold corresponding to the minimum error, the temporary threshold is associated with the classifier. When there are a plurality of temporary threshold values corresponding to the minimum error, the threshold value at the time before threshold estimation corresponding to the classifier is determined based on the plurality of temporary threshold values. The spread prediction system according to any one of claims 20 to 22, further comprising: a function estimation unit that estimates a threshold function from a threshold corresponding to a pre-estimation time and a spread rate. The spread prediction system according to any one of claims 20 to 22, wherein the test data label determination unit calculates a threshold using a threshold function input from outside the spread prediction system. When the customer starts using the product or service, the use start information indicates the use start time of the product or service, and when the product or service is not used, the use start information indicating that the product or service is not used, and the use start Use a customer database that stores customer data, which is a set of individual data for each customer, including one or more items representing customer attributes other than information, and use the product or service at the current time that determines the time to be judged Generate test data that includes individual data for customers who have not started,
Label individual data of customers who have started using the goods or services at the current time, label second data to individual data of customers who have not started using the goods or services, This is data in which the number of individual data labeled with the first label is changed according to the degree of influence, which is the degree that the customer of the product or service urges others to use the product or service within a certain period. Generate learning data,
Based on the learning data, a classifier that is a rule for determining which one of the first label and the second label to label the individual data of the customer from the item representing the attribute of the customer,
A spread prediction method, wherein a label for each individual data in the test data is determined from the classifier and each individual data item in the test data. For each piece of data that is determined to be labeled with the first label, information indicating the use start time is given,
Update the current time increased by a certain amount of time to a new current time,
When the current time is updated, test data at the current time after the update is generated,
When the current time is updated, learning data at the updated current time is generated,
When the current time is updated and new learning data is generated, a new classifier is generated based on the learning data,
The label for each individual data in the test data is determined from the classifier and each individual data item in the test data when the current time is updated and new test data is generated. The spread forecast method described. When the customer starts using the product or service, the use start information indicates the use start time of the product or service, and when the product or service is not used, the use start information indicating that the product or service is not used, and the use start Use a customer database that stores customer data, which is a set of individual data for each customer, including one or more items representing customer attributes other than information, and use the product or service at the current time that determines the time to be judged Generate test data that includes individual data for customers who have not started,
The first label is labeled on individual data of customers who have started using the product or service at the current time, and the second label is labeled on individual data of customers who have not started using the product or service. Generating labeling data, calculating a ratio of the number of individual data labeled with the first label in the labeling data as a penetration rate of the product or service,
Based on the labeling data, a classifier that is a rule for determining a probability score that the first label is labeled on the individual data of the customer from the item representing the customer attribute,
The threshold value is calculated by substituting the calculated penetration rate into a threshold function that is a function of the threshold value of the score with respect to the penetration rate, and each individual item in the test data is calculated from the classifier and each individual data item in the test data. A score of probability that the first label is labeled on the data is determined, and among the individual data in the test data, it is determined that the label for the individual data having a score equal to or higher than the threshold is the first label, and the score is the threshold It determines that the label with respect to less than individual data is a 2nd label, The penetration prediction method characterized by the above-mentioned. For each piece of data that is determined to be labeled with the first label, information indicating the use start time is given,
Update the current time increased by a certain amount of time to a new current time,
When the current time is updated, test data at the current time after the update is generated,
When updating the current time, generate the labeling data at the current time after the update, calculate the penetration rate in the labeling data,
When new labeling data is generated by updating the current time, a new classifier is generated based on the labeling data,
When the current time is updated and new test data is generated, it is confirmed that the first label is labeled on each individual data in the test data from the classifier and each individual data item in the test data. A uniqueness score is determined, a new threshold value is calculated by substituting the calculated penetration rate, and a label for individual data having a score equal to or higher than the threshold value among the individual data in the test data is determined to be the first label. The spread prediction method according to claim 27, wherein a label for individual data having a score less than the threshold is determined to be a second label. When the customer starts using the product or service, the use start information indicates the use start time of the product or service, and when the product or service is not used, the use start information indicating that the product or service is not used, and the use start A spread prediction program installed in a computer including a customer database that stores customer data that is a set of individual data for each customer including one or more items representing customer attributes other than information,
In the computer,
Using the customer database, test data generation processing for generating test data that is data including individual data of customers who have not started using goods or services at the current time for determining the determination target time;
Label individual data of customers who have started using the goods or services at the current time, label second data to individual data of customers who have not started using the goods or services, This is data in which the number of individual data labeled with the first label is changed according to the degree of influence, which is the degree that the customer of the product or service urges others to use the product or service within a certain period. Learning data generation process for generating learning data,
A classifier generation process for generating a classifier that is a rule for determining which of the first label and the second label to label the individual data of the customer from the item representing the attribute of the customer based on the learning data; and,
The spread prediction program for performing the test data label determination process which determines the label with respect to each individual data in the said test data from the said classifier and the item of each individual data in the said test data. In the computer,
In the test data label determination process, information indicating the use start time is given to the individual data determined to be labeled with the first label,
Executing a current time update process in which the current time is increased by a certain amount of time and the new current time is set,
When the current time is updated, the test data generation process generates test data at the current time after the update,
When the current time is updated, the learning data generation process generates learning data at the updated current time,
When new learning data is generated by updating the current time, a new classifier is generated based on the learning data in the classifier generation process,
When new test data is generated by updating the current time, a label for each individual data in the test data from the classifier and each individual data item in the test data in the test data label determination process The spread prediction program according to claim 29. When the customer starts using the product or service, the use start information indicates the use start time of the product or service, and when the product or service is not used, the use start information indicating that the product or service is not used, and the use start A spread prediction program installed in a computer including a customer database that stores customer data that is a set of individual data for each customer including one or more items representing customer attributes other than information,
In the computer,
Using the customer database, test data generation processing for generating test data that is data including individual data of customers who have not started using goods or services at the current time for determining the determination target time;
The first label is labeled on individual data of customers who have started using the product or service at the current time, and the second label is labeled on individual data of customers who have not started using the product or service. Labeling data generation processing for generating labeling data and calculating a ratio of the number of individual data labeled with the first label in the labeling data as a penetration rate of the product or service;
A classifier generating process for generating a classifier that is a rule for determining a probability score that the first label is labeled on the individual data of the customer from the item representing the attribute of the customer based on the labeling data;
By substituting the penetration rate calculated in the labeling data generation process into a threshold function that is a function of the threshold value of the score with respect to the penetration rate, a threshold value is calculated, and from the individual data items in the classifier and the test data, the A score of the probability that the first label is labeled to each individual data in the test data is determined, and it is determined that the label for the individual data in which the score is equal to or higher than the threshold among the individual data in the test data is the first label And a spread prediction program for executing a test data label determination process for determining that a label for individual data having a score less than the threshold is the second label. In the computer,
In the test data label determination process, information indicating the use start time is given to the individual data determined to be labeled with the first label,
Executing a current time update process in which the current time is increased by a certain amount of time and the new current time is set,
When the current time is updated, the test data generation process generates test data at the current time after the update,
When the current time is updated, the labeling data generation process generates labeling data at the current time after the update, and calculates the diffusion rate in the labeling data,
When new labeling data is generated by updating the current time, a new classifier is generated based on the labeling data in the classifier generation process,
When the current time is updated and new test data is generated, the test data label determination process changes the classifier and each individual data item in the test data to each individual data in the test data. A score of probability that one label is labeled is determined, a threshold value is newly calculated by substituting the penetration rate calculated in the labeling data generation process into the threshold function, and the score among the individual data in the test data is The spread prediction program according to claim 31, wherein a label for individual data that is equal to or greater than a threshold value is determined to be a first label, and a label for individual data having a score less than the threshold value is determined to be a second label. Specify the current time used as an estimation target time of the degree of influence, which is the degree to which the customer of the product or service urges others to use the product or service within a certain period, and a certain time before the current time And a plurality of temporary influence levels that are candidates for influence level are input, and a previous time that is a time before the time interval is calculated from the current time,
When the customer starts using the product or service, the use start information indicates the use start time of the product or service, and when the product or service is not used, the use start information indicating that the product or service is not used, and the use start Using a customer database that stores customer data that is a set of individual data for each customer including one or more items representing customer attributes other than information, and starting to use the product or service at the current time Generating current time data by labeling individual data of existing customers with a first label and labeling individual data of customers who have not started using the goods or services with a second label;
For each temporary influence degree, the first label is labeled on the individual data of the customer who has started using the product or service at the previous time, and the individual data of the customer who has not started using the product or service Labeling the second label, generating the previous time data which is data obtained by changing the number of individual data labeled the first label according to the temporary influence degree,
Based on the previous time, a classifier that is a rule for determining whether to label individual customer data from the item representing the customer attribute for each individual previous time data. Generate
For each classifier generated for each previous time data, predict the label that is labeled on the individual data in the current time data from the classifier and the individual data items in the current time data, and predict Calculate the error between the result and the current time data,
Among the errors calculated for each classifier, the smallest error is specified, and when there is one temporary influence corresponding to the smallest error, the temporary influence is determined as the influence, An impact estimation method, wherein when a plurality of temporary impacts corresponding to an error exist, the impact is determined based on the plurality of temporary impacts. 34. The current time data is generated from a set of individual data excluding individual data of customers who have started to use goods or services at the previous time among the customer data stored in the customer database. Degree estimation method. When the customer starts using the product or service, the use start information indicates the use start time of the product or service, and when the product or service is not used, the use start information indicating that the product or service is not used, and the use start An impact estimation program installed in a computer including a customer database for storing customer data, which is a set of individual data for each customer including one or more items representing customer attributes other than information,
A current time used as an influence degree estimation target time, a time interval for designating a predetermined time before the current time, and a plurality of temporary influence degrees as influence degree candidates, are input from the current time to the time Calculating a previous time which is a time before the interval, and using the customer database, a first label is labeled on individual data of a customer who has started using the product or service at the current time, and the product or service Current time data generation processing for generating current time data in which the second label is labeled on individual data of customers who have not started using
For each temporary influence degree, the first label is labeled on the individual data of the customer who has started using the product or service at the previous time, and the individual data of the customer who has not started using the product or service A previous time data group generation process for generating the previous time data which is data obtained by labeling the second label and changing the number of individual data labeled with the first label according to the temporary influence degree;
Based on the previous time, a classifier that is a rule for determining whether to label individual customer data from the item representing the customer attribute for each individual previous time data. Generation process of the previous time classifier group to be generated,
For each classifier generated for each previous time data, predict the label that is labeled on the individual data in the current time data from the classifier and the individual data items in the current time data, and predict An error group calculation process for calculating an error between the result and the current time data; and
Among the errors calculated for each classifier, the smallest error is specified, and when there is one temporary influence corresponding to the smallest error, the temporary influence is determined as the influence, An influence degree estimation program for executing an influence degree calculation process for determining an influence degree based on the plurality of provisional influence degrees when there are a plurality of provisional influence degrees corresponding to errors. In the computer,
In the current time data generation process, the current time data is generated from a set of individual data excluding the customer data stored in the customer database excluding the individual data of the customer who started using the product or service at the previous time. 36. The influence degree estimation program according to claim 35.

Images