Research On Recommendation Algorithm Based On Bidirectional Feature Of Knowledge Graph

Because of the sparsity of user-item interaction and the problem of cold start in traditional recommendation systems,the recommendation effect is not good.Many scholars try to add side information into the recommendation system to improve the performance of recommendation.The application of the knowledge graph as a kind of side information into the recommendation systems can effectively alleviate the above problems and can improve the accuracy,diversity,and interpretability of recommendations.Therefore,the knowledge graph-based recommendation algorithm has attracted wide attention from scholars in recent years.There are two main subjects in the recommendation system: user and item.Knowledge graphs contain a wealth of knowledge information,which can be used to characterize users and items.The richer the feature embedding of users and items,the better the recommendation effect will be.However,most recommendation methods based on the knowledge graph only consider the application of the knowledge graph in either item feature embedding or user feature embedding.Therefore,based on the existing methods,this paper proposes methods that can expand the features of user and item simultaneously by using the rich structural information and semantic information of the knowledge graph.The main research contents of this article are as follows:(1)A structure-based knowledge graph bidirectional feature recommendation method is proposed.The idea of the method is that,based on extracting the user’s features by the outward propagation method,the process of extracting the item’s features by the inward aggregation method is added.The model can use the structural information of the knowledge graph to expand both the user features and the item features.(2)A structure and feature-based knowledge graph bidirectional feature hybrid recommendation method is proposed.The idea of the method is that,based on extracting the item’s features by connection module and knowledge graph embedding,the process of extracting user’s features by the outward propagation method in the structure-based method is added.The model uses simultaneously the feature information and the structure information of the knowledge graph to extract the features of the item and user respectively.The proposed methods make full use of the structure information and feature information of the knowledge graph.They not only alleviate the sparsity and cold start problem but also achieves the purpose of bidirectional feature expansion of the recommendation system,which can predict accurately the probability of users clicking on items.The two methods are applied to movie,book,and music datasets to compute the predicted AUC and ACC.The proposed models are compared with several state-of-the-art baselines,and the results show that the performance of the proposed models exceeds the advanced baselines.