Content recommendation method based on graph neural network and language model

Abstract

The invention discloses a content recommendation method based on a graph neural network and a language model. The method comprises the following steps: acquiring standard configuration file data and standard interaction data of a user; constructing user interaction graph data; obtaining a preference reasoning result of the user; converting the preference reasoning result of the user into query embedding, and performing sub-graph extraction on the user interaction graph data by using the query embedding to obtain user preference sub-graph data; obtaining intention reasoning sub-graph data of the user; and reasoning the sub-graph data and the graph neural network based on the intention of the user, and obtaining a scoring result of the to-be-recommended content. According to the method, various data sources such as the user configuration file, the interaction graph and the interaction information are combined, the information of user preferences can be more comprehensively captured through the natural language processing technology and the graph neural network, the defect that an existing method depends on explicit data is overcome, potential interests and hobbies of the user can be found out from implicit data, and the user experience is improved. And the recommendation accuracy and coverage are improved.

Description

Technical Field

[0001] The present invention relates to the field of content recommendation methods, and in particular to a content recommendation method based on graph neural network and language model. Background Art

[0002] In the modern Internet environment, content recommendation methods have become an important technology and are widely used in e-commerce, social media, online news and other fields. The goal of the recommendation system is to recommend content that users may be interested in based on their historical behaviors and preferences. Among them, graph-based methods have been widely used in recommendation systems because they can effectively capture the complex relationship between users and content. At the same time, graph neural networks (GNNs) are deep learning models specifically designed to process graph-structured data, which can effectively propagate information between nodes through message passing mechanisms. In addition, natural language processing (NLP) technology has also made significant progress in understanding and generating human language, and can be used to process user profiles and interaction information. Existing recommendation system technologies mainly include collaborative filtering-based methods, feature-based methods, and hybrid model-based methods. Among them, collaborative filtering methods rely on historical interaction data between users and content, and make recommendations by calculating the similarity between users and content. Feature methods make recommendations based on feature information of users and content, such as user demographic information, content tags and descriptions, etc. Hybrid model methods combine collaborative filtering and feature methods to make recommendations using multiple data sources. In graph-based methods, recommendation methods usually construct user-content interaction graphs and use graph embedding techniques to embed users and content into high-dimensional space for similarity calculation and recommendation. Some methods also use graph neural networks to propagate information between users and content, thereby improving the accuracy of recommendations. Although existing recommendation system technologies have solved the recommendation problem to a certain extent, there are still some challenges and limitations. First, existing methods often rely on explicit data of users and content, such as purchase records, ratings, etc., which may contain noise and missing values, affecting the effectiveness of recommendations. Secondly, when existing methods process large-scale data, computational efficiency and scalability become an important issue. In addition, existing recommendation methods usually lack the ability to capture users' implicit preferences and are difficult to meet the diverse needs of users. Finally, existing recommendation methods also have deficiencies in explainability and transparency, making it difficult to explain the reasons behind the recommendation results to users. Summary of the invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a content recommendation method based on graph neural network and language model.

[0004] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0005] A content recommendation method based on graph neural network and language model, comprising the following steps

[0006] S1. Collecting user profile data and interaction data, and preprocessing the user profile data and interaction data to obtain the user's standard profile data and standard interaction data;

[0007] S2. constructing user interaction graph data based on the user's standard profile data and standard interaction data;

[0008] S3, based on the user's standard profile data and standard interaction data, obtaining the user's natural language instruction data, and obtaining the user's preference inference result based on the user's natural language instruction data and the language model;

[0009] S4, converting the user's preference reasoning results into query embedding, and using the query embedding to perform subgraph extraction on the user interaction graph data to obtain user preference subgraph data;

[0010] S5, summarizing the user preference subgraph data into a summary in natural language form, and combining the summary in natural language form with the user's standard profile data, standard interaction data, and user preference subgraph data to obtain the user's intention reasoning subgraph data;

[0011] S6. Based on the user's intention inference subgraph data and graph neural network, obtain the scoring results of the content to be recommended.

[0012] Furthermore, in step S1, the user's profile data includes personal profile data filled in when the user registers; the user's interaction data includes purchase records and ratings.

[0013] Furthermore, in step S2, based on the user's standard profile data and standard interaction data, user interaction graph data is constructed. The specific process is: the user and the content to be recommended are respectively used as two types of nodes in the graph, and the standard interaction data is used as the edge connecting the nodes in the graph. The user's standard profile data and the content to be recommended information are imported into the nodes for combination to construct the user interaction graph data.

[0014] Furthermore, in step S4, query embedding is used to extract subgraphs from the user interaction graph data to obtain auxiliary retrieval data of user preferences. The specific process is: inner product calculation is performed on the query embedding and the edges in the user interaction graph data to obtain a sorting result, and subgraphs related to the user preferences are extracted based on the sorting result to obtain auxiliary retrieval data of the user preferences.

[0015] Furthermore, in step S6, based on the user's intention reasoning subgraph data and graph neural network, the scoring result of the content to be recommended is obtained. The specific process is: embedding the user's intention reasoning subgraph data and the content to be recommended into a high-dimensional space to obtain an embedding vector, and performing message passing through the graph neural network to update the embedding vector, and calculating the inner product between the user embedding and the content embedding to obtain the scoring result of the content to be recommended.

[0016] The present invention has the following beneficial effects:

[0017] (1) Comprehensive utilization of multiple data sources: The present invention combines multiple data sources such as user profiles, interaction graphs, and association information. Through natural language processing technology and graph neural networks, it can more comprehensively capture information about user preferences. This not only makes up for the shortcomings of existing methods that rely on explicit data, but also can discover users' potential interests and hobbies from implicit data, thereby improving the accuracy and coverage of recommendations.

[0018] (2) Improved computing efficiency and scalability: The present invention adopts a phased retrieval framework, which reduces the amount of computing and improves computing efficiency by gradually narrowing the retrieval scope. In particular, when processing large-scale data, the framework of the present invention has better scalability, can quickly respond to user requests, and improve user experience;

[0019] (3) Enhanced explainability and transparency: The present invention uses natural language processing technology to convert user profiles and interaction information into natural language, making the recommendation results more transparent and explainable. Users can clearly understand the reasons behind the recommendation results, which increases the credibility of the recommendation system and user satisfaction.

[0020] (4) Innovative combination of LLM and GNN: This paper combines large language models (LLM) and graph neural networks (GNN) for the first time and applies them to recommendation systems, making full use of the reasoning ability of LLM and the information dissemination ability of GNN to achieve accurate capture of user preferences and intelligent content recommendation. This innovative combination has important breakthroughs and practical value in the field of recommendation systems.

[0021] (5) Improving recommendation diversity: The framework of the present invention can provide diverse recommendation results through multi-stage retrieval and information extraction of different granularities to meet the diverse needs of users. This is of great significance for improving personalized recommendations and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Schematic diagram of the process of a content recommendation method based on graph neural network and language model. DETAILED DESCRIPTION

[0023] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0024] like Figure 1 As shown, a content recommendation method based on graph neural network and language model includes steps S1-S6, which are as follows:

[0025] S1. Collecting user profile data and interaction data, and preprocessing the user profile data and interaction data to obtain the user's standard profile data and standard interaction data.

[0026] In an optional embodiment of the present invention, the present invention needs to collect user profiles and interaction information. The user profile includes personal profile data filled in by the user when registering. The user's interaction data includes purchase records and ratings. Then, the present invention cleans and preprocesses these data to remove noise and missing values ​​to obtain standard data.

[0027] S2. Construct user interaction graph data based on the user's standard profile data and standard interaction data.

[0028] In an optional embodiment of the present invention, the present invention uses users and content to be recommended as two types of nodes in the graph, standard interaction data as edges connecting the nodes in the graph, and imports user standard profile data and content information to be recommended into the nodes for combination to construct user interaction graph data.

[0029] S3. Based on the user's standard profile data and standard interaction data, obtain the user's natural language instruction data, and obtain the user's preference inference result based on the user's natural language instruction data and language model.

[0030] In an optional embodiment of the present invention, the present invention combines the user's profile information into natural language instructions and inputs them into the LLM, uses the reasoning ability and stored knowledge of the LLM to infer the user's preference information, and outputs it in the form of natural language. For example, input: "30-year-old female, likes fashion and food", output: "The user may be interested in fashion clothing and food content".

[0031] S3. Convert the user preference reasoning results into query embeddings, and use the query embeddings to perform subgraph extraction on the user interaction graph data to obtain the user preference subgraph data.

[0032] In an optional embodiment of the present invention, the present invention converts the inferred preference information into a query embedding, and uses it to help extract a subgraph from the interaction graph. Specifically, the present invention calculates the inner product between the query embedding and the edges in the interaction graph to obtain a ranking result, and extracts the subgraph most relevant to the user's preference based on the ranking result.

[0033] S5. Summarize the user preference subgraph data into a summary in natural language form, and combine the summary in natural language form with the user's standard profile data, standard interaction data, and user preference subgraph data to obtain the user's intention reasoning subgraph data.

[0034] In an optional embodiment of the present invention, the present invention further summarizes the extracted subgraph into natural language and combines the summary with the user's purchase history for LLM-based intention reasoning. For example, if the content in the subgraph is mostly fashion clothing, it is inferred that the user may have a shopping intention.

[0035] S6. Based on the user's intention inference subgraph data and graph neural network, obtain the scoring results of the content to be recommended.

[0036] In an optional embodiment of the present invention, the present invention passes the final subgraph to the GNN for message passing, and scores the content for recommendation through the inner product of the user / content embedding. Specifically, the present invention embeds the users and content in the subgraph into a high-dimensional space, and passes messages through the GNN to update the embedding vector. Then, the present invention calculates the inner product between the user embedding and the content embedding to obtain the content score.

[0037] 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.

[0038] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0039] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0040] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0041] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A content recommendation method based on graph neural network and language model, characterized in that: The following steps are included S1. Collecting user profile data and interaction data, and preprocessing the user profile data and interaction data to obtain the user's standard profile data and standard interaction data; S2. constructing user interaction graph data based on the user's standard profile data and standard interaction data; S3, based on the user's standard profile data and standard interaction data, obtaining the user's natural language instruction data, and obtaining the user's preference inference result based on the user's natural language instruction data and the language model; S4, converting the user's preference reasoning results into query embedding, and using the query embedding to perform subgraph extraction on the user interaction graph data to obtain user preference subgraph data; S5, summarizing the user preference subgraph data into a summary in natural language form, and combining the summary in natural language form with the user's standard profile data, standard interaction data, and user preference subgraph data to obtain the user's intention reasoning subgraph data; S6. Based on the user's intention inference subgraph data and graph neural network, obtain the scoring results of the content to be recommended.

2. The content recommendation method based on graph neural network and language model according to claim 1, characterized in that: In step S1, the user's profile data includes the personal profile data filled in when the user registers; the user's interaction data includes purchase records and ratings.

3. The content recommendation method based on graph neural network and language model according to claim 1, characterized in that: In step S2, based on the user's standard profile data and standard interaction data, user interaction graph data is constructed. The specific process is: the user and the content to be recommended are respectively used as two types of nodes in the graph, and the standard interaction data is used as the edge connecting the nodes in the graph. The user's standard profile data and the information of the content to be recommended are imported into the nodes for combination to construct the user interaction graph data.

4. The content recommendation method based on graph neural network and language model according to claim 1, characterized in that: In step S4, query embedding is used to extract subgraphs from the user interaction graph data to obtain auxiliary retrieval data of user preferences. The specific process is: inner product calculation is performed on the query embedding and the edges in the user interaction graph data to obtain a sorting result, and subgraphs related to the user preferences are extracted based on the sorting result to obtain auxiliary retrieval data of the user preferences.

5. The content recommendation method based on graph neural network and language model according to claim 1, characterized in that: In step S6, based on the user's intention reasoning subgraph data and graph neural network, the scoring result of the content to be recommended is obtained. The specific process is: embed the user's intention reasoning subgraph data and the content to be recommended into a high-dimensional space to obtain an embedding vector, and perform message passing through the graph neural network to update the embedding vector, and calculate the inner product between the user embedding and the content embedding to obtain the scoring result of the content to be recommended.

Images