Based On Graph Embedding Representation Article Recommendation Model

The purpose of recommendation system is to reduce the time it takes for users to find information of their own interest from massive data and to achieve personalized recommendations for different users.Article recommendation is an important application in recommender systems.Existing article recommendation models only convert high-dimensional data into low-dimensional data by simply embedding when dealing with sparse data.Therefore,user information and article information are not fully utilized in the process of embedding,resulting in inaccurate model recommendation.Therefore,this thesis proposes an article recommendation model based on graph embedding representation..In order to solve the problem of insufficient use of article feature information and article node association information,this thesis proposes a method of mining article information by dual-channel encoder.The graph auto-encoder and auto-encoder are used to obtain the low-dimensional embeddings of article association features and node features respectively,which not only pays attention to the association features between articles,but also obtains the features of article nodes to avoid feature waste and get better feature representation.Through dual-channel encoder dimensionality reduction and feature extraction,high-dimensional data is converted into low-dimensional data,sparse features are converted into dense features,and the problems of dimensional disaster and data sparseness are avoided.To solve the problem of not making full use of interactive information and implicit relationship between users and articles,this thesis constructs bipartite graph to model user information.After obtaining the "user-article" association matrix,after embedding the low-dimensional representation of the association matrix,the probability matrix decomposition is used to update the low-dimensional embedding of the "user-article" association matrix,article node features and article association features,and finally complete recommend.Experiments show that the model in this thesis is superior to other recommendation models,can effectively improve the quality of recommendation,and can complete tasks such as user preference prediction and personalized recommendation.Finally,the practical application proves the effectiveness of the model.