Research And Implementation Of Session-based Recommendation Based On Knowledge Graph

With the rapid development of the Mobile Internet and Big Data,information service resources have been exploded.How to quickly and accurately obtain the most valuable information service resources from massive data has become a huge challenge.Because the session-based recommendation system does not require user history information to generate personalized recommendations that match the user's current preferences based on the user's session sequence,it has received extensived attention and research.As a kind of information representation of related entities in the real world,the knowledge graph can supplement the related information of session items and generate more accurate recommendation.However,there is no related application research and implementation of session-based recommendation with knowledge graphs.And current session-based recommendation still has the following problems:1)Strict sequence modeling,ignoring the weak sequence constraint of the session,without considering the semantic information of items and the external context of session;2)it does not consider the semantic relationship between the items,and cannot be deep mining complex interactions between items.In view of the above problems,this paper carries out the research on the extraction of the external context of session,the complex relationship modeling between session items and the application of session-based recommendation with knowledge graph.The main works are as follows:(1)Design and implement a session-based recommendation method with context aware based on knowledge graph.The external context of the session is construct by knowledge graph and combined with GRU sequence modeling to generate recommendations.On Movielens dataset,the Recall and MRR are improved by 4.36%and 1.34%respectively compared to the baseline model.(2)Design and implement an improved SR-GNN method.Completing the directed session graph of the session base on the link information between the graph entities,and using the gated graph neural network in combination with the multi-head attention mechanism to generate recommendations.On Movielens dataset,the Recall and MRR are improved by 2.89%and 1.21%respectively compared to SR-GNN.(3)Design and implement the information search recommendation system platform for the film field.Complete the search,browse,session-baed recommendation function of the system platform,and implement the application of the session-based recommendation based on knowledge graph model prosed in this paper.