Research On Global Session Recommendation Model Based On Knowledge Graph

Recommender systems play an important role in mitigating the problem of explosive growth of information.Among them,session-based recommendation has the ability to capture the dynamic evolution of users’ interests,and has become an important method to help people find items that may be of interest.Using knowledge graph as auxiliary information can provide auxiliary information very well.There are the following problems in the current session recommendation based on knowledge graph:(1)The knowledge graph is processed as a homogeneous information network,which loses the heterogeneity and rich semantic information of the knowledge graph.(2)The association between the knowledge graph and sequence of interactive items in the session is not considered,the complex interactive relationship between items cannot be deeply excavated.In response to the above problems,this paper focuses on the heterogeneity processing of knowledge graphs in session recommendation and the complex transformation relationship of items in sessions.Firstly,through the analysis of item portraits and person portraits in the recommendation,a global session framework based on knowledge graph is proposed,which uses knowledge graph as auxiliary information,and uses complex items conversion between sessions to enhance the characterization of item portraits.By considering the interaction between items The dynamic portraits of the characters are depicted in sequence.Secondly,aiming at the problem of knowledge graph information processing of items in session recommendation,an item embedding algorithm based on knowledge graph heterogeneity is proposed.The algorithm filters the knowledge graph and then introduces it into the interactive project.By projecting the heterogeneous entities in the knowledge graph into the same entity space,and applying the graph convolutional network to aggregate the projected entities to calculate the embedding of the entities,the heterogeneous knowledge graph is mined.Entities have rich semantic information for recommended items.Thirdly,in view of the complex conversion relationship of items in sessions,a session item embedding algorithm is proposed,which obtains the association between items in session recommendation by constructing a global graph,and combines with knowledge graph information to obtain item information.At the same time,a global session recommendation algorithm based on knowledge graph is designed,which can predict the next click item by obtaining the unified representation of the user.Finally,the model proposed in this paper is verified on the Movie Lens public dataset,MRR and Recall are used to evaluate the recommendation performance.