CN110851651A - Personalized video recommendation method and system - Google Patents
Personalized video recommendation method and system Download PDFInfo
- Publication number
- CN110851651A CN110851651A CN201911088257.6A CN201911088257A CN110851651A CN 110851651 A CN110851651 A CN 110851651A CN 201911088257 A CN201911088257 A CN 201911088257A CN 110851651 A CN110851651 A CN 110851651A
- Authority
- CN
- China
- Prior art keywords
- video
- user
- vector
- recommendation
- faiss
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/73—Querying
- G06F16/735—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/783—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/7847—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using low-level visual features of the video content
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Library & Information Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a personalized video recommendation method and a system, wherein the method comprises the following steps: s1, constructing a user vector and a video vector, and storing the vectors in Faiss; s2, judging whether the user needs to search for interest, if yes, executing a step S3, and if not, executing a step S4; s3, updating the user vector based on the user vector of the user and the IDs of friends or users in the same region; and S4, recalling the video in Faiss based on the user vector, and generating a recommended video for the user. The method and the device can still maintain high recommendation efficiency while avoiding the problem of fixed recommended content types in the video recommendation process. Meanwhile, the recommendation is carried out by combining the social relationship network and the characteristics of the social relationship network, and the recommendation effect is good.
Description
Technical Field
The invention relates to the field of content recommendation, in particular to a personalized video recommendation method and system.
Background
With the popularity of various applications, enterprises can collect more and more comprehensive user data, and how to utilize such data to increase revenue is a problem faced by each enterprise. The most common way is to personalize recommendations, especially in an e-commerce, video website, or other content platform. Social networking based recommendations are a common recommendation method. The more likely it is that people of the same or similar minds like the same content, so social networking based recommendations are based on the content liked by friends who focus on them. However, in the recommendation process based on the social network, only the social relationship among the users is considered, but the characteristics of the users are not considered, and the preference of the people in the social network is completely depended.
In addition, aiming at the problems that the type of content recommended by the existing personalized recommendation tends to be unchanged and other interests of a user cannot be continuously explored, the existing recommendation system provides a Multi Armed Bandit (abbreviated as Bandit) algorithm, and the effect of the Bandit algorithm is usually measured by adopting cumulative regret, specifically:
recommended revenue user clicks and not clicks, i.e., not 0 or 1, i.e., Bernoulli revenue, where wB(i) Is the expected yield of the selected middle arm at the i-th trial, w*Is the best of all arms and T is the number of trials.
The method selects the optimal scheme by accumulation regret, and comprises the following specific steps:
1) supposing that the probability distribution of the probability p conforms to the reel (winds, lots) distribution, and 2 wins, lots are participated;
2) maintaining beta distribution parameters of each corresponding type, wherein each experiment has one plus win and one plus no gains adds one;
3) and when recommending, generating a random number b according to the beta distribution of each class each time, and selecting the random number with the maximum value in all the classes to recommend.
The algorithm assigns a Beta distribution to the good (video/article) instead of a single value. And a sorting value is randomly obtained by sampling each time of sorting, and the variability of commodity sorting is enhanced by the randomness. But simultaneously the mean value of beta distribution can change along with the performance of commodity, just so can distinguish the commodity, let the good commodity of performance have a greater probability to obtain high rank value, and can not be like the exposure probability of evenly distributed every commodity and keep unchangeable forever.
However, the Bandit algorithm selects a recommended result by accumulating user performance, and when the data of the type is small when the data is used up at the beginning, the reliability of the result is not high, so that the effect is not good at the initial stage of the algorithm; in addition, when the searched categories are too many, the random number of each category needs to be calculated every time, which consumes much time; each category of each user needs to maintain one beta distribution parameter, and the storage consumption is large.
Therefore, how to overcome the disadvantages of the existing personalized recommendation and realize the personalized video recommendation with high efficiency and low consumption is a problem to be solved in the field.
Disclosure of Invention
The invention aims to provide a personalized video recommendation method and system aiming at the defects of the prior art. By randomly updating the user vector, the content types recommended in the video recommendation process are diversified, and high recommendation efficiency can be maintained. In addition, the system overhead is small when the method is combined with the existing recall algorithm.
In order to achieve the purpose, the invention adopts the following technical scheme:
a personalized video recommendation method, comprising:
s1, constructing a user vector and a video vector, and storing the vectors in Faiss;
s2, judging whether the user needs to search for interest, if yes, executing a step S3, and if not, executing a step S4;
s3, updating the user vector based on the user vector of the user and the IDs of friends or users in the same region;
and S4, recalling the video in Faiss based on the user vector, and generating a recommended video for the user.
Further, the step S1 is specifically:
s1.1, representing the relevant attributes of a user and a video by adopting characteristics;
s1.2, training and generating a feature vector of a user and a video based on a factorization machine;
and S1.3, summing the feature vectors to generate a user vector and a video vector.
Further, whether interest exploration is needed or not is judged by calculating a probability random number.
Further, a random number which is evenly distributed between 0 and 1 is generated, interest exploration is carried out when the generated random number is more than or equal to 0.4, and otherwise, interest exploration is not carried out.
Further, the step S4 is specifically:
calculating by adopting a vector function in Faiss, and performing video recall in Faiss based on a user vector to obtain a video candidate set; sorting the video candidate sets; and adjusting the sorted video candidate set based on the strategy, and returning the recommended video.
The invention also provides a personalized video recommendation system, which comprises:
the vector construction module is used for constructing a user vector and a video vector and storing the vectors in Faiss;
the judging module is used for judging whether the user needs to perform interest exploration or not, if so, the vector updating module is called, and if not, the recommending module is called;
the vector updating module is used for updating the user vector based on the user vector of the user and the IDs of friends or users in the same region;
and the recommendation module is used for recalling the video in Faiss based on the user vector and generating a recommendation video for the user.
Further, the vector construction module comprises:
the characteristic representation module is used for representing the relevant attributes of the user and the video by adopting characteristics;
the characteristic vector generating module is used for generating the characteristic vectors of the user and the video based on the training of the factorization machine;
and the feature vector summation module is used for summing the feature vectors to generate a user vector and a video vector.
Further, whether interest exploration is needed or not is judged by calculating a probability random number.
Further, a random number which is evenly distributed between 0 and 1 is generated, interest exploration is carried out when the generated random number is more than or equal to 0.4, and otherwise, interest exploration is not carried out.
Further, the recommendation module includes:
calculating by adopting a vector function in Faiss, and performing video recall in Faiss based on a user vector to obtain a video candidate set; sorting the video candidate sets; and adjusting the sorted video candidate set based on the strategy, and returning the recommended video.
Compared with the prior art, the invention has the following advantages:
(1) the present invention does not explore interest for each user request. But rather by means of calculating probabilistic random numbers. The problem of fixed recommended content types in the video recommendation process is avoided, and high recommendation efficiency can be maintained;
(2) the method combines the characteristics of the user and the friend characteristics of the relationship network to recommend, and solves the problem that the existing method based on the social relationship network only depends on the social relationship network and does not consider the characteristics of the user;
(3) the method can be combined with the conventional daily recall algorithm, and the quick recall of the video is realized without additionally increasing the system overhead;
(4) the method is based on FM to carry out feature combination, and the feature vectors of users and videos are constructed, so that the calculation complexity is low, but the effect is greatly improved; meanwhile, the method is suitable for large-scale sparse characteristic application environments, and the generalization capability is strong;
(5) the method stores the user vector and the video vector in the Faiss, and carries out operation based on the vector function carried in the Faiss to realize the quick recall of the video.
Drawings
FIG. 1 is a flowchart of a method for personalized video recommendation provided in accordance with an embodiment;
fig. 2 is a system structure diagram of personalized video recommendation provided in the second embodiment.
Detailed Description
The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict.
It should be noted that the drawings provided in the following embodiments are only for illustrating the basic idea of the present invention, and the components related to the present invention are only shown in the drawings rather than drawn according to the number, shape and size of the components in actual implementation, and the type, quantity and proportion of the components in actual implementation may be changed freely, and the layout of the components may be more complicated.
The invention is further described with reference to the following drawings and specific examples, which are not intended to be limiting.
Example one
As shown in fig. 1, the present embodiment provides a personalized video recommendation method, including:
s1, constructing a user vector and a video vector, and storing the vectors in Faiss;
in order to recommend videos to users, the video recommendation method firstly constructs user vectors and video vectors and carries out video recommendation based on the user vectors and the video vectors. The method specifically comprises the following steps:
s1.1, representing the relevant attributes of a user and a video by adopting characteristics;
the attributes of the user comprise basic attributes, behavior attributes and the like, so that the invention can extract user characteristics representing the user attributes, such as ID characteristics (user name and the like), basic attribute characteristics (such as sex, age, city, hobby and the like), and statistical characteristics (such as one week internet surfing time, video browsing type and the like). Accordingly, the video features used for characterizing the video attributes include ID class features (video address, etc.), basic attribute class features (content, presentation location, etc.), statistical class features (video click-through rate, viewing time, etc.). The feature representation of the user or the video may be extracted directly from an existing image system, or may be extracted from the features of the user or the video to perform the feature representation, and is not limited herein.
S1.2, training and generating a feature vector of a user and a video based on a factorization machine;
the user and the video comprise a plurality of characteristics, and the association exists between different characteristics, for example, a male user prefers a ball game video and a female prefers an entertainment video. Therefore, the feature combination is extremely important in the recommendation sorting process, however, the feature combination amount is large in the recommendation process, and if manual feature combination is performed, the workload is large. Therefore, a Factorization Machine (FM) proposed by redle in 2010 is widely used by various large-scale factories in the field of CTR (click-through rate) estimation and recommendation, and the FM concept solves the workload of a large amount of artificial feature combination.
The second order FM model is specifically:
where n is the number of sample features, xi、xjThe values of the i-th and j-th dimension features, ω0、ωiAre model parameters.In the form of a generic linear model,are cross terms, i.e., combinations of features. v. ofi、vjAre hidden vectors of the ith and jth dimension features respectively,<·,·>is the dot product of vectors, the length of the hidden vector isK factors describing the characteristics are included.
The FM model introduces a second-order feature combination of any two features, FM learns a one-dimensional vector of size k for each feature, and then two features xiAnd xjBy a vector v corresponding to the featureiAnd vjInner product. And the two features do not appear in the training example at the same time, and a new feature combination can be obtained through the inner product of the two feature vectors, so that the method is better in use in a large-scale sparse feature application environment and strong in generalization capability.
therefore, the quadratic term of FM is simplified to V onlyi,fIn relation to the method, the FM can predict new samples in linear time, the complexity is the same as that of a common linear model, and the effect is greatly improved.
The FM model learns the model parameters and the weight vector of each feature during training. Therefore, for the user and the video, a batch of feature weight values, namely feature vectors, of the user and the video are generated based on the training of the factorization machine. The FM model can effectively learn the model parameter, omegaiAnd ωjDoes not require feature x for updatingiAnd xjAnd appears in one sample. The FM model of the invention performs low-dimensional continuous space conversion on sparse feature combinations, and features xiAnd xjModel parameters can still be updated without the occurrence of a sample, so that the feature combination method based on the FM model can be applied to the feature subset { x }i,xjSparse case.
And S1.3, summing the feature vectors to generate a user vector and a video vector.
A user vector,The video vector includes a plurality of feature vector sets. Assuming that the feature set includes n features, the user vector, video vector may be represented as { x }1,x2,...,xnIn which xiRepresenting each of the n feature vectors.
The video recommendation process for the user mainly comprises a recall stage and a sorting stage, wherein the recall stage is to select a part of a video set as a candidate set and calculate based on a user vector and a video vector. However, during a video recall, a large number of candidate sets may need to be recalled, and thus, a large amount of processing may be required for recommendations for a single user. For example, when the recommendation system needs to recall 5000 video candidates during the recall phase, if the vector representations of the users and videos are both 32-dimensional, the computation process that needs to be performed for a single recommendation for a single user is 5000 × 32 — 512 ten thousand. Such a large vector computation amount cannot be tolerated for conventional indexing. Therefore, the user vector and the video vector are stored in the Faiss, and the operation is carried out based on the vector function carried in the Faiss, so that the quick recall of the video is realized.
S2, judging whether the user needs to search for interest, if yes, executing a step S3, and if not, executing a step S4;
when the user sends a request, the corresponding video is recommended for the user, and the sent request can be the access to a video website and the like. In the video recommendation process, if recommendation is performed based on fixed information, such as user preferences, the recommended content types are relatively fixed, and other interests of the user cannot be continuously explored. Therefore, in order to avoid the problem of fixed recommended content types in the video recommendation process and maintain high recommendation efficiency, the invention does not perform interest exploration for each user request. But rather by means of calculating probabilistic random numbers.
For example, a random number uniformly distributed between 0 and 1 is generated, interest exploration is performed when the generated random number is greater than or equal to 0.4, and interest exploration is not performed otherwise.
S3, updating the user vector based on the user vector of the user and the IDs of friends or users in the same region;
when interest exploration is performed, the vector of the user needs to be updated. Specifically, when the user request is obtained, a user vector is first constructed according to information such as the user ID, and the specific steps are consistent with step S1, which is not described herein again. And when interest exploration is carried out, replacing the ID class characteristics of the user with the IDs of friends or users in the same region to obtain a new user vector. Therefore, when interest exploration is carried out, the invention combines the information of the friends and the users in the same region and expands the interest of the users according to the interests of the friends and the users in the same region. In addition, in the interest exploration process, only the ID class characteristics are replaced, and the basic attribute class, the statistical class characteristics and the like of the user are reserved, so that the method combines the characteristics of the user and the friend characteristics of the relationship network to recommend, and solves the problem that the existing method based on the social relationship network recommendation only depends on the social relationship network and does not consider the characteristics of the user.
And S4, recalling the video in Faiss based on the user vector, and generating a recommended video for the user.
The personalized video recommendation mainly comprises three parts: video recall, ranking, and policy adjustment. The video recall means that a part of videos in a large amount of videos are selected as a candidate set. The sorting is mainly to adjust the display order of the videos in the recalled candidate set, for example, sorting the videos in the candidate set from high to low based on the click rate. After sorting, some policy adjustments are made, such as controlling the number of videos of the same uploading author, etc.
According to the method, the user vectors and the video vectors in the sample set are stored in the Faiss, so that after a user sends a request, if the user needs to search, the video recall is carried out in the Faiss based on the updated user vectors, and the calculation process of the recall is operated based on the vector function carried in the Faiss, so that the quick recall of the video is realized. And if the user does not need to explore, the user vector is not updated, and the vector constructed based on the user information is directly recalled.
The method does not limit the specific recall algorithm, can be combined with the conventional daily recall algorithm, does not need to additionally increase the system overhead, and realizes the quick recall of the video.
Example two
As shown in fig. 2, the present embodiment provides a personalized video recommendation system, including:
the vector construction module is used for constructing a user vector and a video vector and storing the vectors in Faiss;
in order to recommend videos to users, the video recommendation method firstly constructs user vectors and video vectors and carries out video recommendation based on the user vectors and the video vectors. The method specifically comprises the following steps:
the characteristic representation module is used for representing the relevant attributes of the user and the video by adopting characteristics;
the attributes of the user comprise basic attributes, behavior attributes and the like, so that the invention can extract user characteristics representing the user attributes, such as ID characteristics (user name and the like), basic attribute characteristics (such as sex, age, city, hobby and the like), and statistical characteristics (such as one week internet surfing time, video browsing type and the like). Accordingly, the video features used for characterizing the video attributes include ID class features (video address, etc.), basic attribute class features (content, presentation location, etc.), statistical class features (video click-through rate, viewing time, etc.). The feature representation of the user or the video may be extracted directly from an existing image system, or may be extracted from the features of the user or the video to perform the feature representation, and is not limited herein.
The characteristic vector generating module is used for generating the characteristic vectors of the user and the video based on the training of the factorization machine;
the user and the video comprise a plurality of characteristics, and the association exists between different characteristics, for example, a male user prefers a ball game video and a female prefers an entertainment video. Therefore, the feature combination is extremely important in the recommendation sorting process, however, the feature combination amount is large in the recommendation process, and if manual feature combination is performed, the workload is large. Therefore, a Factorization Machine (FM) proposed by redle in 2010 is widely used by various large-scale factories in the field of CTR (click-through rate) estimation and recommendation, and the FM concept solves the workload of a large amount of artificial feature combination.
The second order FM model is specifically:
where n is the number of sample features, xi、xjThe values of the i-th and j-th dimension features, ω0、ωiAre model parameters.In the form of a generic linear model,are cross terms, i.e., combinations of features. v. ofi、vjAre hidden vectors of the ith and jth dimension features respectively,<·,·>is the dot product of vectors, the length of the hidden vector isK factors describing the characteristics are included.
The FM model introduces a second-order feature combination of any two features, FM learns a one-dimensional vector of size k for each feature, and then two features xiAnd xjBy a vector v corresponding to the featureiAnd vjInner product. And the two features do not appear in the training example at the same time, and a new feature combination can be obtained through the inner product of the two feature vectors, so that the method is better in use in a large-scale sparse feature application environment and strong in generalization capability.
therefore, the quadratic term of FM is simplified to V onlyi,fIn relation to the method, the FM can predict new samples in linear time, the complexity is the same as that of a common linear model, and the effect is greatly improved.
The FM model learns the model parameters and the weight vector of each feature during training. Therefore, for the user and the video, a batch of feature weight values, namely feature vectors, of the user and the video are generated based on the training of the factorization machine. The FM model can effectively learn the model parameter, omegaiAnd ωjDoes not require feature x for updatingiAnd xjAnd appears in one sample. The FM model of the invention performs low-dimensional continuous space conversion on sparse feature combinations, and features xiAnd xjModel parameters can still be updated without the occurrence of a sample, so that the feature combination method based on the FM model can be applied to the feature subset { x }i,xjSparse case.
And the feature vector summation module is used for summing the feature vectors to generate a user vector and a video vector.
The user vector and the video vector comprise a plurality of feature vector sets. Assuming that the feature set includes n features, the user vector, video vector may be represented as { x }1,x2,...,xnIn which xiRepresenting each of the n feature vectors.
The video recommendation process for the user mainly comprises a recall stage and a sorting stage, wherein the recall stage is to select a part of a video set as a candidate set and calculate based on a user vector and a video vector. However, during a video recall, a large number of candidate sets may need to be recalled, and thus, a large amount of processing may be required for recommendations for a single user. For example, when the recommendation system needs to recall 5000 video candidates during the recall phase, if the vector representations of the users and videos are both 32-dimensional, the computation process that needs to be performed for a single recommendation for a single user is 5000 × 32 — 512 ten thousand. Such a large vector computation amount cannot be tolerated for conventional indexing. Therefore, the user vector and the video vector are stored in the Faiss, and the operation is carried out based on the vector function carried in the Faiss, so that the quick recall of the video is realized.
The judging module is used for judging whether the user needs to perform interest exploration or not, if so, the vector updating module is called, and if not, the recommending module is called;
when the user sends a request, the corresponding video is recommended for the user, and the sent request can be the access to a video website and the like. In the video recommendation process, if recommendation is performed based on fixed information, such as user preferences, the recommended content types are relatively fixed, and other interests of the user cannot be continuously explored. Therefore, in order to avoid the problem of fixed recommended content types in the video recommendation process and maintain high recommendation efficiency, the invention does not perform interest exploration for each user request. But rather by means of calculating probabilistic random numbers.
For example, a random number uniformly distributed between 0 and 1 is generated, interest exploration is performed when the generated random number is greater than or equal to 0.4, and interest exploration is not performed otherwise.
The vector updating module is used for updating the user vector based on the user vector of the user and the IDs of friends or users in the same region;
when interest exploration is performed, the vector of the user needs to be updated. Specifically, when the user request is obtained, a user vector is first constructed according to information such as the user ID, and the specific steps are consistent with step S1, which is not described herein again. And when interest exploration is carried out, replacing the ID class characteristics of the user with the IDs of friends or users in the same region to obtain a new user vector. Therefore, when interest exploration is carried out, the invention combines the information of the friends and the users in the same region and expands the interest of the users according to the interests of the friends and the users in the same region. In addition, in the interest exploration process, only the ID class characteristics are replaced, and the basic attribute class, the statistical class characteristics and the like of the user are reserved, so that the method combines the characteristics of the user and the friend characteristics of the relationship network to recommend, and solves the problem that the existing method based on the social relationship network recommendation only depends on the social relationship network and does not consider the characteristics of the user.
And the recommendation module is used for recalling the video in Faiss based on the user vector and generating a recommendation video for the user.
The personalized video recommendation mainly comprises three parts: video recall, ranking, and policy adjustment. The video recall means that a part of videos in a large amount of videos are selected as a candidate set. The sorting is mainly to adjust the display order of the videos in the recalled candidate set, for example, sorting the videos in the candidate set from high to low based on the click rate. After sorting, some policy adjustments are made, such as controlling the number of videos of the same uploading author, etc.
According to the method, the user vectors and the video vectors in the sample set are stored in the Faiss, so that after a user sends a request, if the user needs to search, the video recall is carried out in the Faiss based on the updated user vectors, and the calculation process of the recall is operated based on the vector function carried in the Faiss, so that the quick recall of the video is realized. And if the user does not need to explore, the user vector is not updated, and the vector constructed based on the user information is directly recalled.
The method does not limit the specific recall algorithm, can be combined with the conventional daily recall algorithm, does not need to additionally increase the system overhead, and realizes the quick recall of the video.
Therefore, the personalized recommendation method and the personalized recommendation system provided by the invention do not perform interest exploration for each request of the user. But rather by means of calculating probabilistic random numbers. The problem of fixed recommended content types in the video recommendation process is avoided, and high recommendation efficiency can be maintained; the method combines the characteristics of the user and the friend characteristics of the relationship network to recommend, and solves the problem that the existing method based on the social relationship network only depends on the social relationship network and does not consider the characteristics of the user; meanwhile, the method can be combined with the conventional daily recall algorithm, and the quick recall of the video is realized without additionally increasing the system overhead; in addition, feature combination is carried out based on FM, feature vectors of users and videos are constructed, the calculation complexity is low, and the effect is greatly improved; meanwhile, the method is suitable for large-scale sparse characteristic application environments, and the generalization capability is strong; and finally, storing the user vector and the video vector in Faiss, and performing operation based on a vector function carried in the Faiss to realize quick recall of the video.
It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims (10)
1. A method for personalized video recommendation, comprising:
s1, constructing a user vector and a video vector, and storing the vectors in Faiss;
s2, judging whether the user needs to search for interest, if yes, executing a step S3, and if not, executing a step S4;
s3, updating the user vector based on the user vector of the user and the IDs of friends or users in the same region;
and S4, recalling the video in Faiss based on the user vector, and generating a recommended video for the user.
2. The personalized video recommendation method according to claim 1, wherein the step S1 specifically comprises:
s1.1, representing the relevant attributes of a user and a video by adopting characteristics;
s1.2, training and generating a feature vector of a user and a video based on a factorization machine;
and S1.3, summing the feature vectors to generate a user vector and a video vector.
3. The method of claim 1, wherein whether interest exploration is required is determined by calculating a probabilistic random number.
4. The personalized video recommendation method of claim 3, wherein a random number is generated that is evenly distributed between 0 and 1, and interest exploration is performed when the generated random number is greater than or equal to 0.4, and not performed otherwise.
5. The personalized video recommendation method according to claim 3, wherein the step S4 specifically comprises:
calculating by adopting a vector function in Faiss, and performing video recall in Faiss based on a user vector to obtain a video candidate set; sorting the video candidate sets; and adjusting the sorted video candidate set based on the strategy, and returning the recommended video.
6. A personalized video recommendation system, comprising:
the vector construction module is used for constructing a user vector and a video vector and storing the vectors in Faiss;
the judging module is used for judging whether the user needs to perform interest exploration or not, if so, the vector updating module is called, and if not, the recommending module is called;
the vector updating module is used for updating the user vector based on the user vector of the user and the IDs of friends or users in the same region;
and the recommendation module is used for recalling the video in Faiss based on the user vector and generating a recommendation video for the user.
7. The personalized video recommendation system of claim 6, wherein the vector construction module comprises:
the characteristic representation module is used for representing the relevant attributes of the user and the video by adopting characteristics;
the characteristic vector generating module is used for generating the characteristic vectors of the user and the video based on the training of the factorization machine;
and the feature vector summation module is used for summing the feature vectors to generate a user vector and a video vector.
8. The personalized video recommendation system of claim 6, wherein the interest exploration is determined by calculating a probabilistic random number.
9. The personalized video recommendation system of claim 8, wherein a random number is generated that is evenly distributed between 0-1, wherein interest exploration is performed when the generated random number is greater than or equal to 0.4, and wherein interest exploration is not performed otherwise.
10. The personalized video recommendation method of claim 8, wherein the recommendation module comprises:
calculating by adopting a vector function in Faiss, and performing video recall in Faiss based on a user vector to obtain a video candidate set; sorting the video candidate sets; and adjusting the sorted video candidate set based on the strategy, and returning the recommended video.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911088257.6A CN110851651B (en) | 2019-11-08 | 2019-11-08 | Personalized video recommendation method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911088257.6A CN110851651B (en) | 2019-11-08 | 2019-11-08 | Personalized video recommendation method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110851651A true CN110851651A (en) | 2020-02-28 |
CN110851651B CN110851651B (en) | 2022-07-22 |
Family
ID=69599955
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911088257.6A Active CN110851651B (en) | 2019-11-08 | 2019-11-08 | Personalized video recommendation method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110851651B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111444388A (en) * | 2020-03-30 | 2020-07-24 | 杭州趣维科技有限公司 | Video tag sorting method based on random walk |
CN112836085A (en) * | 2021-02-08 | 2021-05-25 | 深圳市欢太科技有限公司 | Weight adjusting method and device and storage medium |
CN112949703A (en) * | 2021-02-23 | 2021-06-11 | 广州市百果园信息技术有限公司 | Video recall method and device and electronic equipment |
CN115018477A (en) * | 2022-08-09 | 2022-09-06 | 大白熊大数据科技(常熟)有限公司 | Big data analysis method and equipment based on enterprise OA system |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102594905A (en) * | 2012-03-07 | 2012-07-18 | 南京邮电大学 | Method for recommending social network position interest points based on scene |
CN103870517A (en) * | 2012-12-09 | 2014-06-18 | 祁勇 | Method and system for acquiring personalized features of user |
CN104008138A (en) * | 2014-05-08 | 2014-08-27 | 南京邮电大学 | Music recommendation method based on social network |
US20170098236A1 (en) * | 2015-10-02 | 2017-04-06 | Yahoo! Inc. | Exploration of real-time advertising decisions |
US20170188102A1 (en) * | 2015-12-23 | 2017-06-29 | Le Holdings (Beijing) Co., Ltd. | Method and electronic device for video content recommendation |
WO2019029046A1 (en) * | 2017-08-11 | 2019-02-14 | 深圳市耐飞科技有限公司 | Video recommendation method and system |
CN109871858A (en) * | 2017-12-05 | 2019-06-11 | 北京京东尚科信息技术有限公司 | Prediction model foundation, object recommendation method and system, equipment and storage medium |
CN110162701A (en) * | 2019-05-10 | 2019-08-23 | 腾讯科技(深圳)有限公司 | Content delivery method, device, computer equipment and storage medium |
-
2019
- 2019-11-08 CN CN201911088257.6A patent/CN110851651B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102594905A (en) * | 2012-03-07 | 2012-07-18 | 南京邮电大学 | Method for recommending social network position interest points based on scene |
CN103870517A (en) * | 2012-12-09 | 2014-06-18 | 祁勇 | Method and system for acquiring personalized features of user |
CN104008138A (en) * | 2014-05-08 | 2014-08-27 | 南京邮电大学 | Music recommendation method based on social network |
US20170098236A1 (en) * | 2015-10-02 | 2017-04-06 | Yahoo! Inc. | Exploration of real-time advertising decisions |
US20170188102A1 (en) * | 2015-12-23 | 2017-06-29 | Le Holdings (Beijing) Co., Ltd. | Method and electronic device for video content recommendation |
WO2019029046A1 (en) * | 2017-08-11 | 2019-02-14 | 深圳市耐飞科技有限公司 | Video recommendation method and system |
CN109871858A (en) * | 2017-12-05 | 2019-06-11 | 北京京东尚科信息技术有限公司 | Prediction model foundation, object recommendation method and system, equipment and storage medium |
CN110162701A (en) * | 2019-05-10 | 2019-08-23 | 腾讯科技(深圳)有限公司 | Content delivery method, device, computer equipment and storage medium |
Non-Patent Citations (3)
Title |
---|
S RENDLE: "Factorization Machines", 《ICDM 2010》 * |
代丽等: "个性化推荐系统综述", 《计算机软件及计算机应用》 * |
张富国: "基于社交网络的个性化推荐技术", 《小型微型计算机系统》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111444388A (en) * | 2020-03-30 | 2020-07-24 | 杭州趣维科技有限公司 | Video tag sorting method based on random walk |
CN111444388B (en) * | 2020-03-30 | 2023-06-30 | 杭州小影创新科技股份有限公司 | Video tag ordering method based on random walk |
CN112836085A (en) * | 2021-02-08 | 2021-05-25 | 深圳市欢太科技有限公司 | Weight adjusting method and device and storage medium |
CN112949703A (en) * | 2021-02-23 | 2021-06-11 | 广州市百果园信息技术有限公司 | Video recall method and device and electronic equipment |
CN112949703B (en) * | 2021-02-23 | 2024-08-23 | 广州市百果园信息技术有限公司 | Video recall method and device and electronic equipment |
CN115018477A (en) * | 2022-08-09 | 2022-09-06 | 大白熊大数据科技(常熟)有限公司 | Big data analysis method and equipment based on enterprise OA system |
Also Published As
Publication number | Publication date |
---|---|
CN110851651B (en) | 2022-07-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Ma et al. | Off-policy learning in two-stage recommender systems | |
Sun et al. | A framework for recommending accurate and diverse items using bayesian graph convolutional neural networks | |
CN110162693B (en) | Information recommendation method and server | |
CN110851651B (en) | Personalized video recommendation method and system | |
Xu | A novel recommendation method based on social network using matrix factorization technique | |
Guan et al. | Matrix factorization with rating completion: An enhanced SVD model for collaborative filtering recommender systems | |
CN110321422B (en) | Method for training model on line, pushing method, device and equipment | |
Zhao et al. | Interactive collaborative filtering | |
US9900358B1 (en) | Organizing a stream of content | |
Elena et al. | Survey of multiarmed bandit algorithms applied to recommendation systems | |
CN106651544B (en) | Conversational recommendation system with minimal user interaction | |
CN108960945A (en) | Method of Commodity Recommendation and device | |
TW201447797A (en) | Method and system for multi-phase ranking for content personalization | |
US11636394B2 (en) | Differentiable user-item co-clustering | |
WO2009023067A1 (en) | System and method for invitation targeting in a web-based social network | |
US11048764B2 (en) | Managing under—and over-represented content topics in content pools | |
WO2012009180A1 (en) | Value maximizing recommendation systems | |
CN110413894A (en) | The training method of commending contents model, the method for commending contents and relevant apparatus | |
Wu et al. | Optimization matrix factorization recommendation algorithm based on rating centrality | |
US10984058B2 (en) | Online diverse set generation from partial-click feedback | |
CN110472016A (en) | Article recommended method, device, electronic equipment and storage medium | |
Akhil et al. | A survey of recommender system types and its classification. | |
Zhang et al. | Improving augmented reality using recommender systems | |
CN118467851B (en) | Artificial intelligent data searching and distributing method and system | |
CN113836393B (en) | Cold start recommendation method based on preference self-adaptive meta-learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB02 | Change of applicant information |
Address after: 22nd floor, block a, Huaxing Times Square, 478 Wensan Road, Xihu District, Hangzhou, Zhejiang 310000 Applicant after: Hangzhou Xiaoying Innovation Technology Co.,Ltd. Address before: 16 / F, HANGGANG Metallurgical Science and technology building, 294 Tianmushan Road, Xihu District, Hangzhou City, Zhejiang Province, 310012 Applicant before: HANGZHOU QUWEI SCIENCE & TECHNOLOGY Co.,Ltd. |
|
CB02 | Change of applicant information | ||
GR01 | Patent grant | ||
GR01 | Patent grant |