JP5506867B2 - Content distribution device - Google Patents

Description

  The present invention relates to a technique for distributing content such as advertisements via a network such as the Internet.

  In Internet homepages and the like, there are often advertisement columns for display advertisements typified by banner advertisements.

  Some relatively high-priced advertising products with high advertising effectiveness have user groups selected by behavioral targeting and area targeting methods as inventory, and there are those that guarantee a predetermined number of advertisements (impressions) to those user groups. .

  On the other hand, in Patent Document 1, an advertisement is selected based on a judgment of suitability between a search character string based on an action designated by a search and an action character string in which an action to be displayed for each advertisement is associated in advance. A technique for realizing an effective search-linked advertisement that matches a searcher's search intention is disclosed. Statistical index is used in the generation of the search character string.

  Patent Document 2 discloses a technology for determining content to be distributed by statistical analysis by collecting logs of terminals that are actually displayed in the distribution of rotation advertisements and estimating an estimated click rate by multiple regression analysis or the like. It is disclosed.

  Patent Document 3 discloses a technique for sending a message to one or more servers by a browser script and polling the one or more servers for an auction of an ad impression displayed on a web page.

JP 2012-37920 A JP 2011-243125 A JP 2010-514073 A

  As mentioned above, advertising products based on impression guarantee based on behavioral targeting and area targeting have already been realized, but there is a limit to the effectiveness of advertising for specific campaigns such as replacement of mobile phones and purchase of home appliances and cars. . This is because conventional behavioral targeting and area targeting determine whether or not they are interested in a specific field based on user characteristics, so actual behavior such as replacement of mobile phones, purchase of home appliances and cars, etc. It is because it is not analyzed in connection with.

  The present invention has been proposed in view of the above-described conventional problems, and the object of the present invention is to appropriately extract a group of impression target users for advertisements of a specific campaign and to increase the value of advertising products. It is in.

  In order to solve the above-mentioned problem, in the present invention, a means for identifying a user who has performed an action targeted for a specific campaign from log data, and extracting action content within a predetermined period of the user as correct answer data; Means for extracting, as random data, action contents within a predetermined period from randomly selected users other than the user who is the target of the correct data from the log data, and the target of the specific campaign from the correct data and the random data Means for extracting action content having a high correlation with the action to be obtained as a feature quantity item; from the log data based on the feature quantity item; a feature quantity of a user that is a target of the correct answer data; and a target of the random data And generating the feature quantity of the user, and the feature quantity and the target of the specific campaign from the feature quantity A means for calculating a weighting coefficient indicating a relationship with movement, a means for generating a feature quantity for all users from the log data based on the feature quantity item, and a user-specific score from the weighting coefficient and the feature quantity for all users Means for calculating a top predetermined number of the user-specific scores as a target user list, means for presenting impressions to the target user list and executing an auction for the specific campaign, Means for delivering the contents of the winning bidder to the users included in the target user list.

  In the content distribution apparatus of the present invention, it is possible to appropriately extract a user group as an impression target for an advertisement or the like of a specific campaign, and to increase the value of an advertising product.

It is a figure which shows the structural example of the system concerning one Embodiment of this invention. It is a flowchart which shows the process example of embodiment. It is a figure which shows the example of log data. It is a figure which shows the example of extraction of campaign information. It is a figure which shows the example of extraction of correct answer data and random data (in the case of the model change of a mobile telephone). It is a figure which shows the example of extraction of correct answer data and random data (in the case of replacement by purchase of a car). It is a figure which shows the example of extraction of a feature-value item. It is a figure which shows the example of the production | generation of a feature-value. It is a figure which shows the example of calculation of a weighting coefficient. It is a figure which shows the example of the production | generation of all the user feature-values. It is a figure which shows the example of calculation of a score classified by user. It is a figure which shows the example of extraction of an advertising object user. It is a figure which shows the example of auction execution and advertisement delivery.

  Hereinafter, preferred embodiments of the present invention will be described.

<Configuration>
FIG. 1 is a diagram showing a configuration example of a system according to an embodiment of the present invention.

  In FIG. 1, the system provides a plurality of terminal devices 2 operated by a general user, a plurality of terminal devices 3 operated by an advertiser user, and an information providing service for a general user using the terminal device 2. And a plurality of service server devices 4 that perform the auction of a campaign (specific campaign) that guarantees the ad impression to the advertiser user who operates the terminal device 3, and distributes the successful bidder's advertisement 5 is connected to a network 1 such as the Internet. Note that the advertisement management device 5 may have a path connected directly or indirectly (including offline data exchange) with the service server device 4.

  The terminal devices 2 and 3, the service server device 4, and the advertisement management device 5 have a hardware configuration of a general computer device.

  The advertisement management apparatus 5 includes a log acquisition unit 501, a campaign information acquisition unit 503, a correct data / random data extraction unit 505, and a feature amount item extraction unit 508 as functional units mainly realized by software (computer program). A feature amount generation unit 510, a weighting coefficient calculation unit 512, an all-user feature amount generation unit 514, a score calculation unit 516, an advertisement target user extraction unit 518, an auction execution unit 520, and an advertisement distribution unit 522. It has.

  The log acquisition unit 501 has a function of acquiring log data 502 from the service server device 4 online or offline.

  The campaign information acquisition unit 503 has a function of acquiring campaign information 504 input and set in advance by the operator, or extracting and acquiring the campaign information 504 from the log data 502.

  Based on the campaign information 504, the correct data / random data extraction unit 505 did not perform the correct data 506, which is information on the user who performed the campaign target action, from the log data 502 and the campaign target action. It has a function of extracting random data 507 that is user information (which could not be determined to have been performed).

  The feature quantity item extraction unit 508 has a function of extracting the feature quantity item 509 constituting the feature quantity from the correct answer data 506 and the random data 507.

  The feature quantity generation unit 510 has a function of generating feature quantities 511 corresponding to individual users of the correct answer data 506 and the random data 507 from the feature quantity item 509, the correct answer data 506, and the random data 507. The random data 507 when generating the feature quantity 511 preferably uses more data than when the feature quantity item 509 is obtained in order to increase the accuracy of later calculation of the weighting coefficient.

  The weighting coefficient calculation unit 512 has a function of calculating the weighting coefficient calculation unit 512 used for calculating a score indicating the possibility that each user performs the action to be the target of the campaign from the feature amount 511.

  The all-user feature value generation unit 514 has a function of generating an all-user feature value 515 including individual feature values of all users within a predetermined range from the feature value item 509 and the log data 502.

  The score calculation unit 516 has a function of calculating a user-specific score 517 based on the weighting coefficient 513 and the total user feature amount 515.

  The advertisement target user extraction unit 518 has a function of extracting a user who is highly likely to perform a campaign target action based on the user-specific score 517 and outputting it as an advertisement target user list 519.

  Based on the campaign information 504, the auction execution unit 520 executes an auction for the campaign that guarantees the advertisement impression to the advertisement target user list 519 for an advertiser who may participate in the campaign, and obtains the successful bid result 521. It has a function to output.

  When the general user included in the advertisement target user list 519 makes a page request to the service server apparatus 4 from the terminal device 2 based on the successful bid result 521 and the advertisement target user list 519, the advertisement distribution unit 522 The content of the advertiser's campaign advertisement is obtained from the advertisement database 523 and distributed.

<Operation>
FIG. 2 is a flowchart showing a processing example of the above embodiment.

  First, in FIG. 2A, the log acquisition unit 501 of the advertisement management apparatus 5 acquires log data 502 from the service server apparatus 4 online or offline regularly or according to an operator's instruction (step S0).

  FIG. 3 is a diagram showing an example of the log data 502. “Cookie information” “User ID” “Device” “Date” “Action content (search keyword, browsing URL, browsing video ID, browsing advertisement ID, product purchase (category) , Manufacturer, product name, price, ..),.

  “Cookie information” is information that is written in the browser when accessing the service server device 4 from the browser of the terminal device 2, and it is possible to specify that access has been made from the same browser unless the browser cookie is reset. . The “user ID” is information for specifying the user of the terminal device 2 and is specified by logging in when the service server device 4 provides a service. “Device” is information indicating the device type of the terminal device 2. When the terminal device 2 is a mobile phone, a model name is recorded. When the terminal device 2 is a PC (Personal Computer), it indicates that the device is a PC or a model. The name is recorded. “Date” is a date when the terminal device 2 accesses the service server device 4. “Action content (search keyword, browsing URL, browsing video ID, browsing advertisement ID, product purchase (category, manufacturer, product name, price,...),...)” Indicates that the terminal device 2 accesses the service server device 4. It is a record of the contents of the action that was taken.

  Next, in FIG. 2B, the campaign information acquisition unit 503 of the advertisement management apparatus 5 acquires the campaign information 504 input and set in advance by the operator, or extracts the campaign information 504 from the log data 502. Obtain (step S1).

  FIG. 4 is a diagram illustrating an example of extracting the campaign information 504 from the log data 502.

  In FIG. 4, the campaign information acquisition unit 503 classifies the contents of browsing / purchasing included in the log data 502 by category and manufacturer, and displays the number of browsing (the total number of browsing for each manufacturer within the category) or sales (within the category). A category in which a predetermined number (for example, five companies) or more of the manufacturers with which the total amount of purchase for each manufacturer antagonizes, and competing manufacturers in each category are output as campaign information 504 (step S11).

  Returning to FIG. 2B, the correct answer data / random data extraction unit 505 of the advertisement management device 5 is based on the campaign information 504, and the correct answer data that is information on the user who performed the campaign target action from the log data 502. 506 and random data 507, which is information of a user who has not performed the action that is the subject of the campaign, are extracted (step S2). Note that the correct answer data 506 and the random data 507 can be generated separately. In this case, the correct answer data 506 and the random data 507 that have already been generated are used in subsequent processing.

  FIGS. 5 and 6 are diagrams showing examples of extraction of correct answer data 506 and random data 507. FIG. 5 shows a case where the campaign is “mobile phone model change” and FIG. 6 shows a case where the campaign is “car purchase”. As an example.

  In FIG. 5, the correct data / random data extraction unit 505 analyzes the log data 502 to change the model of a user whose device has been changed from one mobile phone model to another mobile phone model with the same user ID. The action content from that time to a predetermined period before is extracted as correct answer data 506 (step S21). In the log data 502 shown in FIG. 3, only the same user ID “yid001” is shown. However, on the date “20120306”, the device has changed from the previous mobile phone model “SS900” to the mobile phone model “NN500”. It is determined that the model has been changed.

  Returning to FIG. 5, the correct data / random data extraction unit 505 extracts, as random data 507, action contents from a user other than the correct data to a predetermined period before the current time (step S <b> 22).

  In FIG. 6, the correct data / random data extraction unit 505 analyzes the log data 502 to extract, as correct data 506, action contents from that time to a predetermined period before a user who purchased a car with the same user ID. (Step S23).

  Next, the correct data / random data extraction unit 505 extracts the action content from a point other than the correct data user up to a predetermined period before as the random data 507 (step S24).

  Returning to FIG. 2B, the feature quantity item extraction unit 508 of the advertisement management apparatus 5 extracts the feature quantity item 509 constituting the feature quantity from the correct answer data 506 and the random data 507 (step S3).

  FIG. 7 is a diagram illustrating an example of extraction of the feature amount item 509.

In FIG. 7, the feature quantity item extraction unit 508 cuts out search keywords, browsing URLs, browsing video IDs, browsing advertisement IDs, and the like included in the action content of the correct data 506 as entities e, and correct data 506 and random data 507. The mutual information amount I (a, b _e ) is calculated from (step S31).

U _patt is a set of user selected as the correct data _{506, U rand} is a set of user selected as a random data 507, _{U e} is the set of users who viewed the entity e, u is an individual user, _{p e} the number of users who have viewed the entity e of the user of the correct answer data 506, the r _e is the number of users who have viewed the entity e of the user of the random data 507.

a discrete random variable, b _e indicating whether user u does not belong belong to correct data 506 is a discrete random variable that indicates whether no viewing, the user u views the entity e.

Here, probabilities p (a, b _e ) for all combinations of the discrete random variables a and b _e are calculated as shown in the table. Note that p (a) is indicated in the case of the sum (Sum) in the case of b _e = 1 and b _e = 0. In the case of a sum (Sum) when a = 1 and a = 0, p (b _e ) is indicated. N is the total number of users, N _patt is the number of users in the correct answer data 506, N _rand is the number of users in the random data 507, and the sum of N _patt and N _rand is N. n _e is the number of users who browse the entity e, the sum of _{p e} and _{r e} is a _{n e.}

From these, the mutual information I (a, b _e ) is calculated by the formula shown in the figure.

Next, the feature quantity item extraction unit 508 outputs a predetermined number of entities e having large mutual information I (a, b _e ) as the feature quantity item 509 (step S32).

  Returning to FIG. 2B, the feature amount generation unit 510 of the advertisement management apparatus 5 uses the feature amount item 509, the correct answer data 506, and the random data 507 to correspond to the individual users of the correct answer data 506 and the random data 507. A quantity 511 is generated (step S4). For the random data 507, it is preferable to use more samples than when the feature amount item 509 is calculated.

  FIG. 8 is a diagram illustrating an example of generation of the feature amount 511.

  In FIG. 8, the feature quantity generation unit 510 determines whether or not each entity has acted on the correct answer data 506 and the random data 507 for the feature quantity data having the number of bits corresponding to the number of entities in the feature quantity item 509. A feature value 511 is generated for each user by representing “1” and “0” (step S41).

  Returning to FIG. 2B, the weighting coefficient calculation unit 512 of the advertisement management apparatus 5 includes the weighting coefficient calculation unit 512 for calculating a score indicating the possibility that each user performs an action targeted for the campaign. It calculates from 511 (step S5).

  FIG. 9 is a diagram illustrating an example of calculating the weighting coefficient 513.

  In FIG. 9, the weighting coefficient calculation unit 512 obtains the result of the inner product of the weighting coefficient vector (w1, w2,...) And the feature quantity vector (vector having each bit as an element). The random data is set to “0” (step S51), and the weighting coefficient 513 is determined by regression analysis (logistic regression analysis) (step S52). For example, the weighting coefficient 513 is determined by an operation using the least square method or a convergence operation (iterative operation).

  Returning to FIG. 2B, the all-user feature value generation unit 514 of the advertisement management apparatus 5 calculates the all-user feature value 515 corresponding to each of all users within a predetermined range from the feature value item 509 and the log data 502. Generate (step S6).

  FIG. 10 is a diagram showing an example of generation of all user feature values 515.

  In FIG. 10, the all-user feature quantity generation unit 514 performs the behavior of each entity based on the log data 502 from the present time to a predetermined period before the feature quantity data having the number of bits corresponding to the number of entities in the feature quantity item 509. By expressing “1” or “0” as to whether or not the user has performed, a feature amount is generated for each user and output as all user feature amounts 515 (step S61).

  Returning to FIG. 2B, the score calculation unit 516 of the advertisement management apparatus 5 calculates the user-specific score 517 based on the weighting coefficient 513 and the total user feature amount 515 (step S7).

  FIG. 11 is a diagram illustrating an example of calculation of the user-specific score 517.

  In FIG. 11, the score calculation unit 516 calculates the score for each user by taking the inner product of the vector of the weighting coefficient 513 and the vector of the individual feature values of all the user feature values 515, and outputs the score as a user-specific score 517 (step S71).

  Returning to FIG. 2B, the advertisement target user extraction unit 518 of the advertisement management apparatus 5 extracts a user who is likely to perform an action that is the target of the campaign based on the user-specific score 517, and is an advertisement target user list. It outputs as 519 (step S8).

  FIG. 12 is a diagram illustrating an example of extraction of advertisement target users.

  In FIG. 12, the advertisement target user extraction unit 518 sorts the user-specific scores 517 in descending order of score, and outputs a predetermined upper number of users as the advertisement target user list 519 (step S81).

  Returning to FIG. 2B, the auction execution unit 520 of the advertisement management apparatus 5 performs a campaign auction that guarantees an advertisement impression on the advertisement target user list 519 to an advertiser who may participate in the campaign. The advertisement distribution unit 522 executes a page request when a general user included in the advertisement target user list 519 makes a page request to the service server device 4 based on the successful bid result 521 and the advertisement target user list 519. The content of a certain advertiser's campaign advertisement is acquired from the advertisement database 523 and distributed (step S9).

  FIG. 13 is a diagram illustrating an example of auction execution and advertisement distribution.

  In FIG. 13, the auction execution unit 520 gives a campaign content (for example, change of mobile phone model, purchase of a car, etc.), impression (for example, a user in the user list to be advertised) to competing manufacturers for the adopted campaign information 504 The number is 1000, 3 times per user, the impression is 3000), etc., and a request for participation in the auction is made (step S91).

  Next, the auction execution unit 520 makes a successful bid to the maker who has presented the maximum amount by the deadline type auction and outputs a successful bid result 521 (step S92).

  Next, the operator or the manufacturer (advertiser) who has made a successful bid creates or selects the content of the campaign advertisement for which the bid has been made (step S93).

  Next, the advertisement distribution unit 522 distributes the content of the campaign advertisement in response to the advertisement distribution request from the service server device 4 in response to the page request to the service server device 4 from the user included in the advertisement target user list 519 ( Step S94).

<Summary>
As described above, according to the present embodiment, it is possible to appropriately extract the impression target user group for the advertisement of the specific campaign and the like, and it is possible to increase the value of the advertising product.
The present invention has been described above by the preferred embodiments of the present invention. While the invention has been described with reference to specific embodiments, various modifications and changes may be made to the embodiments without departing from the broad spirit and scope of the invention as defined in the claims. Obviously you can. In other words, the present invention should not be construed as being limited by the details of the specific examples and the accompanying drawings.

DESCRIPTION OF SYMBOLS 1 Network 2, 3 Terminal device 4 Service server apparatus 5 Advertisement management apparatus 501 Log acquisition part 502 Log data 503 Campaign information acquisition part 504 Campaign information 505 Correct answer / random data extraction part 506 Correct answer data 507 Random data 508 Feature-value item extraction part 509 Feature quantity item 510 Feature quantity generation unit 511 Feature quantity 512 Weighting coefficient calculation section 513 Weighting coefficient 514 All user feature quantity generation section 515 All user feature quantity 516 Score calculation section 517 User specific score 518 Advertisement target user extraction section 519 Advertisement target user List 520 Auction execution part 521 Successful bid result 522 Advertisement distribution part 523 Advertisement database

Claims (6)

A means for identifying a user who has performed an action that is a target of a specific campaign from log data, and extracting action contents within a predetermined period of the user as correct data;
Means for extracting, as random data, action contents within a predetermined period from users randomly selected other than the user who is the target of the correct data from the log data;
Means for extracting action contents having a high correlation with the action targeted for the specific campaign from the correct data and the random data as feature quantity items;
Means for generating, from the log data based on the feature quantity item, a feature quantity of a user who is a target of the correct data and a feature quantity of a user who is a target of the random data;
Means for calculating a weighting coefficient indicating a relationship between the feature quantity and an action targeted for the specific campaign from the feature quantity;
Means for generating feature quantities of all users from the log data based on the feature quantity items;
Means for calculating a user-specific score from the weighting coefficient and the feature amount of all users;
Means for extracting a top predetermined number of the user-specific scores as a target user list;
Means for presenting impressions to the target user list and executing an auction for the specific campaign;
A content distribution apparatus comprising: means for distributing content of a successful bidder of the auction to users included in the target user list. The content delivery apparatus according to claim 1,
The action content included in the log data is classified by category and manufacturer, and includes means for extracting a category in which a predetermined number or more of manufacturers with which the number of browsing or sales antagonizes exist,
A content distribution apparatus that determines the specific campaign based on an extracted category. The content distribution apparatus according to claim 2, wherein
A means for extracting competing manufacturers in the extracted category is provided.
A content distribution apparatus that executes an auction for the specific campaign for the extracted manufacturer. From the correct answer data indicating the action content within a predetermined period of the user who performed the action targeted for the specific campaign, and the random data indicating the action content within the predetermined period of the user selected at random other than the user A means of extracting behavioral content highly correlated with the behavior targeted for the campaign as a feature item,
Means for generating, from the log data based on the feature quantity item, a feature quantity of a user who is a target of the correct data and a feature quantity of a user who is a target of the random data;
Means for calculating a weighting coefficient indicating a relationship between the feature quantity and an action targeted for the specific campaign from the feature quantity;
Means for generating feature quantities of all users from the log data based on the feature quantity items;
Means for calculating a user-specific score from the weighting coefficient and the feature amount of all users;
A content distribution apparatus comprising: means for extracting a distribution target user of content related to the specific campaign based on the user-specific score. A step in which the content distribution device identifies a user who has performed an action that is a target of a specific campaign from log data, and extracts the action content of the user within a predetermined period as correct data;
The content distribution device extracts, as random data, action contents within a predetermined period from users randomly selected from the log data other than the user who is the target of the correct data;
The content distribution device extracting from the correct answer data and the random data, action content highly correlated with the action targeted for the specific campaign as a feature quantity item;
The content distribution device generates, from the log data based on the feature amount item, a feature amount of a user that is a target of the correct data and a feature amount of a user that is a target of the random data;
The content distribution device calculates a weighting coefficient indicating a relationship between the feature amount and an action to be a target of the specific campaign from the feature amount;
The content distribution device generates a feature amount of all users from the log data based on the feature amount item;
The content distribution device calculating a user-specific score from the weighting coefficient and the feature amount of all users;
The content distribution device extracts a top predetermined number of the user-specific scores as a target user list;
The content distribution device presenting impressions to the target user list and executing an auction for the specific campaign;
A content distribution method comprising: a step of distributing the content of the successful bidder of the auction to a user included in the target user list. A step in which the content distribution device identifies a user who has performed an action that is a target of a specific campaign from log data, and extracts the action content of the user within a predetermined period as correct data;
The content distribution device extracts, as random data, action contents within a predetermined period from users randomly selected from the log data other than the user who is the target of the correct data;
The content distribution device extracting from the correct answer data and the random data, action content highly correlated with the action targeted for the specific campaign as a feature quantity item;
The content distribution device generates, from the log data based on the feature amount item, a feature amount of a user that is a target of the correct data and a feature amount of a user that is a target of the random data;
The content distribution device calculates a weighting coefficient indicating a relationship between the feature amount and an action to be a target of the specific campaign from the feature amount;
The content distribution device generates a feature amount of all users from the log data based on the feature amount item;
The content distribution device calculating a user-specific score from the weighting coefficient and the feature amount of all users;
The content distribution method comprising: a step of extracting a predetermined upper number of the user-specific scores as a target user list.

Images