Film and television recommendation method and system based on graph neural network

Through the graph neural network model, the diversity relationship between users and film and television resources is solved, and the cold start and intent diversity ignorance in traditional film and television recommendation methods are achieved, more accurate and personalized film and television recommendations are achieved, and user experience and platform competition is improved. force.

CN118939832AInactive Publication Date: 2024-11-12JINAN UNIVERSITY
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202410935761.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-12
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional film and television recommendation methods face cold start problems and ignore the limitations of the diversity of user behavioral intentions and the correlation between film and television resources, resulting in poor recommendation results.

Method used

Using the graph neural network model, users and film and television resources are used as nodes to establish a two-part graph of positive intention and negative intention, extract nonlinear relationships through multi-layer perceptrons, and obtain the interactive score between users and film and television content. Based on this Recommend film and television content for users.

Benefits of technology

It significantly improves the accuracy and user satisfaction of film and television recommendations, provides a more accurate and personalized film and television recommendation experience, and reduces the proportion of negative content users feel in the recommendation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118939832A_ABST
    Figure CN118939832A_ABST
Patent Text Reader

Abstract

The invention discloses a film and television recommendation method and system based on a graph neural network, and belongs to the technical field of computer data mining and analysis, and the method comprises the steps: collecting user interaction data, carrying out the preprocessing, building a graph neural network recommendation model through employing a graph structure, and enabling a user and a film and television to serve as nodes, respectively establishing bipartite graphs of the positive intention and the negative intention of the user; based on a graph neural network recommendation model, a film and television platform data set with score data of a network open source is used as training data to carry out model training, information of negative intention representation is extracted to enhance positive intention representation, a nonlinear relation is enriched for an enhanced result through a multi-layer perceptron, and an enhanced result is obtained. Obtaining interaction scores between the user and different film and television contents; and recommending the film and television content to the user based on the interaction score. According to the method, the diversity of user behavior intentions and the relevance between movie and television resources are fully considered, and more comprehensive and in-depth movie and television recommendation experience is brought to the user.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of computer data mining and analysis, and in particular to a film and television recommendation method and system based on graph neural network. Background Art

[0002] With the rapid development of information technology and the widespread application of the Internet, film and television resources have shown explosive growth. When faced with massive amounts of film and television content, how to quickly find films and television works that suit personal preferences has become a major challenge for users. Traditional film and television recommendation methods mainly rely on user historical behavior data and metadata information of film and television content, but these methods often face the cold start problem, that is, for new users or new film and television resources, due to the lack of sufficient historical data support, their recommendation effects are often unsatisfactory. In addition, traditional methods usually only focus on the static relationship between users and film and television resources, ignoring the diversity of user behavior intentions and the correlation between film and television resources.

[0003] In recent years, the rise of graph neural networks has brought new breakthroughs in the field of film and television recommendations. As a powerful deep learning model, graph neural networks are good at processing graph structured data, can deeply explore the relationship between nodes and edges, and achieve accurate modeling and analysis of complex systems. By mapping users and film and television resources as nodes in the graph, graph neural networks can effectively capture the potential connection between user interests and film and television resources, while considering the diversity of user behavioral intentions, thereby significantly optimizing the effect of film and television recommendations. The present invention intends to design a new technology for film and television recommendations through graph neural networks to improve the accuracy of film and television recommendations and user satisfaction. Summary of the invention

[0004] In order to solve the above problems, the purpose of the present invention is to propose a film and television recommendation technology based on graph neural network, which aims to overcome the situation of ignoring the diversity of user intentions and effectively reduce the proportion and importance of content that users feel negative in the recommendation results by introducing a graph neural network model, thereby improving the accuracy of film and television recommendations and user satisfaction.

[0005] In order to achieve the above technical objectives, the present application provides a method for recommending movies and TV shows based on a graph neural network, comprising the following steps:

[0006] After collecting and preprocessing user interaction data, a graph neural network recommendation model is established using a graph structure, which uses users and movies as nodes to establish a bipartite graph of users' positive and negative intentions respectively;

[0007] Based on the graph neural network recommendation model, the open source film and television platform dataset with rating data is used as training data for model training, and the information of negative intent representation is extracted to enhance the positive intent representation. The multi-layer perceptron is used to enrich the nonlinear relationship of the enhanced result to obtain the interactive ratings between users and different film and television content.

[0008] Recommend film and television content to users based on the interaction score.

[0009] Preferably, in the process of acquiring different interaction intentions, the user interaction data is preprocessed, including mapping of user ID and film and television ID, extracting rating information and dividing interaction behavior intentions.

[0010] Preferably, during the mapping process, the user ID and the film and television ID are mapped to consecutive positive integers through a dictionary.

[0011] Preferably, in the process of dividing the interaction behavior intention, the scoring data in the obtained interaction records is evaluated by setting a scoring threshold, and the interaction records greater than the scoring threshold are regarded as the user's positive intention behavior, and the interaction records less than the scoring threshold are regarded as the user's negative intention behavior.

[0012] Preferably, in the process of establishing the bipartite graph, the bipartite graph of the positive intention behavior and the negative intention behavior is represented as:

[0013] Edge Pos ={(u, i, p)|p∈F pos}

[0014] Edge Neg ={(u, i, p)|p∈F Neg}

[0015] Graph Pos ,Graph Neg = {U, I, Edge Pos}, {U, I, Edge Neg}

[0016] In the formula, F pos is the set of positive intention behaviors of the user, F Neg is the set of negative intention behaviors of users, u represents users, i represents movies, p represents ratings, U represents the set of users in the dataset, I represents the set of movies in the dataset, Edge Pos represents the edge set between users and their high-rated movies. Neg Represents the edge set between users and their low-rated movies, Graph Pos Indicates positive intention, Graph Neg Indicates negative intention.

[0017] Preferably, in the process of obtaining the interaction score, a feature matrix of N user nodes and M film and television nodes of the input graph neural network is constructed;

[0018] Based on the feature matrix, the interaction matrix of positive and negative intents is obtained according to the propagation mode of the graph neural network.

[0019] According to the interaction matrix, the positive intention and negative intention generalization matrices are obtained respectively, and the information of the feature nodes is aggregated. The final features are extracted by the multi-layer perceptron as follows:

[0020] E Pos °=MLP(E Pos +σ(J0int·E Neg ))

[0021] In the formula, E Pos represents the positive intention representation of users and movies, MLP() represents multi-layer perceptron, E Neg It represents the negative intention representation of users and movies, and σ() represents the sigmoid activation function.

[0022] Preferably, in the process of obtaining the interaction score, based on the feature E Pos °, the user's recommendation results are obtained by calculating the cosine similarity and obtaining the interaction score, where the cosine similarity is expressed as:

[0023]

[0024] In the formula, A represents a vector, B represents a vector, and A i represents the value of the i-th dimension of vector A, n represents the dimension of the vector, B i Represents the value of the i-th dimension of vector B.

[0025] The present invention also discloses a film and television recommendation system based on a graph neural network, which is used to implement a film and television recommendation method, including:

[0026] A data collection module, used to collect user interaction data;

[0027] The data processing module is used to pre-process the user interaction data and then establish a graph neural network recommendation model using a graph structure. It is used to use users and movies as nodes to establish a bipartite graph of the user's positive intention and negative intention respectively.

[0028] The interactive scoring module is used to train the model based on the graph neural network recommendation model. It uses the open source film and television platform dataset with rating data as training data to extract information representing negative intent to enhance positive intent representation, and enriches the nonlinear relationship of the enhanced result through a multi-layer perceptron to obtain the interactive ratings between users and different film and television content.

[0029] The film and television recommendation module is used to recommend film and television content to users based on the level of interaction scores.

[0030] Preferably, the data processing module is also used to divide the user interaction data of click behaviors with user rating records on the streaming media platform, and define different interaction intentions of users according to the levels of the ratings, wherein, for the rating data in the obtained interaction records, by setting a rating threshold, the interaction records greater than the rating threshold are regarded as the user's positive intention behavior, and the interaction records less than the rating threshold are regarded as the user's negative intention behavior.

[0031] Preferably, the interaction scoring module is used to extract the negative intent representation by analyzing the intrinsic correlation between two different intentions, enrich the expressive power of the positive intent representation, extract the nonlinear relationship in the positive intent representation through a multi-layer perceptron, and calculate the interaction score between the user and different film and television content based on the obtained results.

[0032] The present invention discloses the following technical effects:

[0033] The present invention solves the limitations of traditional film and television recommendation methods by introducing a graph neural network model, so as to achieve more accurate and efficient film and television recommendations, and improve user experience and platform competitiveness.

[0034] The present invention fully considers the diversity of user behavior intentions and the correlation between film and television resources, and will bring users a more comprehensive and in-depth film and television recommendation experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0036] Figure 1 It is a flow chart of the steps of the method of the present invention.

[0037] Figure 2 It is a schematic diagram of dividing positive intention and negative intention described in an embodiment of the present invention.

[0038] Figure 3 Schematic diagram of a multi-layer perceptron according to an embodiment of the present invention.

[0039] Figure 4 It is a schematic diagram of the graph neural network structure described in an embodiment of the present invention.

[0040] Figure 5It is a schematic diagram of a feature extraction module described in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0042] like Figure 1-5 As shown, the present invention provides a film and television recommendation method based on graph neural network. By introducing the graph neural network model, the accuracy of film and television recommendation and user satisfaction are improved; the method comprehensively considers the diversity of user historical behavior data and user behavior intentions to provide users with more accurate and personalized film and television recommendation services. At the same time, the method can also deeply explore the correlation between film and television resources and recommend more film and television works that meet their preferences to users, thereby optimizing user experience and improving platform operation efficiency, specifically including the following contents:

[0043] 1. Divide the click behaviors with user rating records on the streaming platform, define different user interaction intentions according to the ratings, and use these two parts of data as input data for mapping user model training;

[0044] 2. Set two representations for each user and film and television content to represent different intentions, and design targeted graph neural networks to model the two different intentions respectively;

[0045] 3. By analyzing the intrinsic relationship between two different intents, a feature extraction module is designed to extract negative intent representations, which can be used to enrich the expressive power of positive intent representations;

[0046] 4. The nonlinear relationship in the positive intent representation is extracted through a multi-layer perceptron, and the results are used to calculate the interaction score between the user and different film and television content, and recommend film and television content to users according to the score.

[0047] The present invention can design targeted graph neural network models for different intentions while distinguishing user behavioral intentions, analyze the internal connections of behavioral intentions, extract key information in the representation of negative intentions to enhance the representation of positive intentions, thereby enhancing the expressiveness of the representation of positive intentions while avoiding the misleading of negative intentions. While improving the model recommendation performance, it effectively reduces the proportion and importance of content that users feel negative in the recommendation results, and increases the proportion and importance of content that users are interested in in the recommendation results.

[0048] Embodiment: The present invention provides a method for recommending movies and TV shows based on a graph neural network, which specifically includes the following steps:

[0049] Step S1, preprocessing the user's historical interaction data, including: mapping of user ID and film ID, extracting rating information and dividing interaction behavior intention;

[0050] Step S2, in order to establish a recommendation model using a graph structure, users and movies are used as nodes of the graph, and a bipartite graph is established for the user's positive intention behavior and negative intention behavior respectively;

[0051] Step S3, using an open source film and television platform dataset with rating data as a training dataset, using the mapping method in S2 to model different intention behaviors respectively, designing a feature extraction module representation to extract information for negative intention representation to enhance positive intention representation, and finally using a multi-layer perceptron to enrich the nonlinear relationship of the enhanced result;

[0052] Step S4, using the representation obtained in S3 to calculate the interaction score between the user and different film and television contents, and recommending film and television contents to the user according to the score.

[0053] Specifically, S1. User interaction data preprocessing: The user interaction data preprocessing process mainly includes ID mapping and intent segmentation.

[0054] S1.1, ID mapping:

[0055] For the film and television platform, the user ID and film and television ID are unique values ​​randomly generated by computers. There is no pattern among them, and it is not convenient for the subsequent construction of the graph structure. Therefore, they are mapped into continuous positive integers through a dictionary.

[0056] S1.2, Intent Division:

[0057] For the interaction records obtained, there are score data with a minimum score of 1 and a maximum score of 5. According to the score data, the interaction records with a score greater than or equal to 4 are regarded as the user's positive intention behavior, and the interaction records with a score less than 4 are regarded as the user's negative intention behavior. The working principle diagram is attached. Figure 2 .

[0058] S2. Use users and movies as nodes to build a bipartite graph:

[0059] For the set of positive intention behaviors and the set of negative intention behaviors obtained in step S1, both users and movies are taken as nodes, and the bipartite graph is established by connecting the edges in the following way:

[0060] Edge Pos ={(u, i, p)|p∈F pos}

[0061] Edge Neg ={(u, i, p)|p∈F Neg}

[0062] Graph Pos ,Graph Neg = {U, I, Edge Pos}, {U, I, Edge Neg}

[0063] Among them, F pos is the set of positive intention behaviors of the user, F Neg is the set of negative intention behaviors of users, u represents users, i represents movies, U represents the set of users in the dataset, and I represents the set of movies in the dataset.

[0064] S3. Establish a graph neural network recommendation model:

[0065] An open-source film and television platform dataset with rating data is used as the training dataset. The mapping method in S2 is used to model different intention behaviors respectively. A feature extraction module is designed to extract information of negative intention representation to enhance positive intention representation. Finally, a multi-layer perceptron is used to enrich the nonlinear relationship of the enhanced result.

[0066] S3.1. Input the feature matrix of N user nodes and M movie nodes of the graph neural network, and we have:

[0067]

[0068] Among them, u i , v j ∈R F , F is the feature vector dimension of each node.

[0069] S3.2. Constructing the propagation method of graph neural network for positive intent Graph Pos , we can use Edge Pos Construct the interaction matrix R Pos ∈R N×M , then we can get the adjacency matrix A Pos, the formula is as follows:

[0070]

[0071] Its generalization matrix It can be calculated by the following formula:

[0072]

[0073] Among them, D Pos A Pos The degree matrix of .

[0074] Similarly, we can get:

[0075]

[0076] Where I is the identity matrix.

[0077] The information of feature nodes is aggregated according to the generalization matrices of the two intents. See the attached diagram for the working principle. Figure 4 :

[0078]

[0079] Design feature extraction module Use it to extract E Neg The feature information in E is weighted Pos , there is E Pos +σ(Joint·E Neg ), σ() is the sigmoid activation function, and the working principle diagram is attached Figure 5 ;

[0080] Finally, the final features are extracted through the multi-layer perceptron, that is,

[0081] E Pos °=MLP(E Pos +σ(Joint·E Neg )).

[0082] S4. Recommend film and television content to users:

[0083] According to the feature E obtained from S3 Pos °, by calculating the cosine similarity y u,v =Cosine Similarity u,v Get the user's recommendation results. See the attached working principle diagram Figure 3 . The cosine similarity is calculated as follows:

[0084]

[0085] The present invention provides a film and television recommendation method based on graph neural network, which overcomes the situation of ignoring the diversity of user intentions and effectively reduces the proportion and importance of content that users feel negative in the recommendation results, thereby obtaining recommendation results with better user experience.

[0086] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0087] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0088] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A film and television recommendation method based on graph neural network, characterized in that: The following steps are involved: After collecting and preprocessing user interaction data, a graph neural network recommendation model is established using a graph structure, which uses users and movies as nodes to establish a bipartite graph of users' positive and negative intentions respectively; Based on the graph neural network recommendation model, the open source film and television platform dataset with rating data is used as training data for model training, and the information of negative intent representation is extracted to enhance the positive intent representation. The multi-layer perceptron is used to enrich the nonlinear relationship of the enhanced result to obtain the interactive rating between users and different film and television content; Based on the interaction score, film and television content is recommended to the user.

2. According to claim 1, a method for recommending movies and TV shows based on graph neural network is characterized in that: In the process of acquiring different interaction intentions, the user interaction data is preprocessed, including mapping of user ID and film and television ID, extracting rating information and dividing interaction behavior intention.

3. According to claim 2, a method for recommending movies and TV shows based on graph neural network, characterized in that: During the mapping process, the user ID and the movie ID are mapped to consecutive positive integers through a dictionary.

4. The method for recommending movies and TV shows based on graph neural network according to claim 3, characterized in that: In the process of dividing the interaction behavior intention, the scoring data in the obtained interaction records is evaluated by setting a scoring threshold. The interaction records with a score greater than the scoring threshold are regarded as the user's positive intention behavior, and the interaction records with a score less than the scoring threshold are regarded as the user's negative intention behavior.

5. The method for recommending movies and TV shows based on graph neural network according to claim 4, characterized in that: In the process of building the bipartite graph, the bipartite graph of positive intention behavior and negative intention behavior is represented as: Edge Pos ={(u,i,p)|p∈F pos } Edge Neg ={(u,i,p)|p∈F Neg } Graph Pos ,Graph Neg ={U,I,Edge Pos },{U,I,Edge Neg } In the formula, F pos is the set of positive intention behaviors of the user, F Neg is the set of negative intention behaviors of users, u represents users, i represents movies, p represents ratings, U represents the set of users in the dataset, I represents the set of movies in the dataset, Edge Pos represents the edge set between users and their high-rated movies. Neg Represents the edge set between users and their low-rated movies, Graph Pos Indicates positive intention, Graph Neg Indicates negative intention.

6. The method for recommending movies and TV shows based on graph neural network according to claim 5, characterized in that: In the process of obtaining interactive scores, a feature matrix of N user nodes and M film nodes of the input graph neural network is constructed; Based on the feature matrix, the interaction matrix of positive intent and negative intent is obtained according to the propagation mode of the graph neural network; According to the interaction matrix, the positive intention and negative intention generalization matrices are obtained respectively, and the information of the feature nodes is aggregated. The final features are extracted by the multi-layer perceptron as follows: IS Pos o =MLP(E Pos +σ(Joint·E Neg )) In the formula, E Pos represents the positive intention representation of users and movies, MLP() represents multi-layer perceptron, E Neg It represents the negative intention representation of users and movies, and σ() represents the sigmoid activation function.

7. The method for recommending movies and TV shows based on graph neural network according to claim 6, characterized in that: In the process of obtaining the interaction score, based on feature E Pos o , the user's recommendation result is obtained by calculating the cosine similarity, and the interaction score is obtained, wherein the cosine similarity is expressed as: In the formula, A represents a vector, B represents a vector, and A i represents the value of the i-th dimension of vector A, n represents the dimension of the vector, B i Represents the value of the i-th dimension of vector B.

8. A film and television recommendation system based on graph neural network, characterized in that: include: A data collection module, used to collect user interaction data; A data processing module, which is used to establish a graph neural network recommendation model using a graph structure after preprocessing the user interaction data, and is used to establish a bipartite graph of the user's positive intention and negative intention respectively by taking the user and the film and television as nodes; An interactive scoring module is used to train the model based on the graph neural network recommendation model, using the open source film and television platform dataset with scoring data as training data, extracting information representing negative intent to enhance positive intent representation, and enriching the nonlinear relationship of the enhanced result through a multi-layer perceptron to obtain interactive scores between users and different film and television content; The film and television recommendation module is used to recommend film and television content to users based on the interaction score.

9. A film and television recommendation system based on graph neural network according to claim 8, characterized in that: The data processing module is also used to divide the user interaction data of click behaviors with user rating records on the streaming media platform, and define different user interaction intentions according to the levels of the ratings, wherein, for the rating data in the obtained interaction records, by setting a rating threshold, the interaction records greater than the rating threshold are regarded as the user's positive intention behavior, and the interaction records less than the rating threshold are regarded as the user's negative intention behavior.

10. A film and television recommendation system based on graph neural network according to claim 9, characterized in that: The interaction scoring module is used to extract negative intent representation by analyzing the intrinsic relationship between two different intentions, enrich the expression ability of positive intent representation, extract the nonlinear relationship in the positive intent representation through a multi-layer perceptron, and calculate the interaction score between the user and different film and television content based on the obtained results.

Citation Information

Patent Citations

  • Interactive movie recommendation system and method based on graph neural network and reinforcement learning

    CN114357241A

  • Information recommendation method and device, equipment and storage medium

    CN115270005A

  • Recommendation method based on combination of graph neural network and Transform and applied to intelligent recommendation system

    CN116561424A