Research On Sequence Recommendation Method Based On Knowledge Graph And Attention Mechanism

A recommendation system is a personalized information recommendation system based on user behavior information,which finds the articles that users are interested in the network and recommends them to users.Nowadays,recommendation system has been widely used in various fields of the Internet,such as music,film,micro-video,food recommendation,ecommerce and other fields,and has achieved good results.Recommendation system plays an important role in alleviating the problem of information overload caused by the rapid development of Internet information.It not only brings more convenient and free experience to people,but also brings more profits to many Internet companies.Therefore,exploring better recommendation algorithms has always been a hot research topic.With the development of recommendation algorithms,sequence-based recommendation systems emerge as the times require.Different from traditional recommendation algorithms mining users' static interests,sequence recommendation models dynamic preferences for users' historical interests,and recommends changes in interests.The effect is better than the static traditional recommendation algorithm.However,the following problems need to be solved:First,data sparsity and cold start of sequence recommendation are more serious;Second,the user's long-term interest,short-term interest and the complex relationship within the sequence are not well mined;Third,the expression of user interest is inaccurate;Fourth,the low return of sequential recommendation prediction model.Considering the shortcomings of the existing work,this paper introduces the knowledge graph,takes advantage of the graph neural network and self-attention network,optimizes the internal structure and recommendation method of the sequence model,and solves the above problems.The main research contents are as follows:(1)This paper proposes a sequence recommendation method(SR-KGA)that integrates knowledge graph and self-attention mechanism.SR-KGA can significantly improve the diversity of sequence-based recommendations and ensure the accuracy of recommendations.SR-KGA improved the recommendation algorithm mainly in three aspects.First of all,it improved the representation accuracy of the item by introducing knowledge graph to represent graph with external information.In addition,to improve the representation accuracy of the sequence model,the seq2 seq model is constructed by using the attention mechanism to represent the sequence model,and the changing trend of user interest is represented by a multivector to better represent the diversity of user interest.Finally,Improve the difference between items in the prediction sequence,optimize the loss function with the diversity regularization term,further improve the diversity of recommendation results,and realize personalized recommendation.(2)This paper proposes a self-supervised learning sequence recommendation method(SRKGSSL)based on knowledge graph.This method introduces the knowledge graph,uses the knowledge graph neural network to characterize the items,uses the dual-channel model,uses the attention mechanism and the graph convolutional neural network GCN to simultaneously mine the complex relationships in the sequence information,and conduct joint training.The model was optimized by comparing learning and mutual information maximization to achieve the best recommendation effect.Through experiments on two datasets,the accuracy of the proposed method has better performance.