The Research And Impleentation Of Context-aware Recomender Based On Bayesian Personalized Ranking

With the explosion of Internet information,it is more and more expensive for users to access information.In order to help users to acquire data more conveniently and accurately,the recommender system has emerged.However,traditional recommender system faces some inherent problems.On the one hand,the traditional recommender only focuses on the scoring data between users and items,and does not consider the contextual information;On the other hand,due to the inherent sparsity of the ratings,traditional recommender system is suffering from a serious decline in the quality of the recommendation.In order to improve the quality of recommendation,more contextual information needs to be introduced,which results in context-aware recommender system.A real-world recommender system contains many types of objects and relationships,which constitute a heterogeneous information network.To leverage heterogeneous information network to integrate a variety of information,and provide users with personalized and accurate recommendation services,become a key issue to be solved in this paper.Moveover,the use of a knowledge base in a hybrid recommendation system has drawn increasing attention as the knowledge base provides a wealth of information,including both structured and unstructured data with different semantics.In response to the above problems,the main contribution is divided into three parts:1.Analyzing the limitations of the traditional social recommender,this paper proposes a novel DSBPR for the first time.The method improves the basic preference predictor by introducing similarity information between users and items in heterogeneous information networks and learns the parameters on BPR.2.Based on BDAP,the data preprocessing algorithm and the collaborative filtering algorithm are parallelized and componentized by using a large-scale distributed computing framework.3.By studying how to use the heterogeneous information network in the knowledge base to improve the quality of recommendation service,the paper proposes a recommender framework for knowledge base embedding,by using two algorithms respectively from the structured content and textual content to extract the semantic representation of the entity.