Movie Recommendation Algorithm Based On Knowledge Graph And Preference Propagation

Acquisition of personalized information for each single Internet user becomes a difficult task currently due to the dramatically increase of online content and services,hence recommendation system is proposed.Recommendation system aims at providing personalized services for users and improve the efficiency of information and resources utilization.There are several problem needs to be solve in recommendation system,including sparse matrix and cold start problem.Therefore,side information,such as social network and item attributes is introduced to improve the cold start of system and enhance the performance.The recommendation algorithm used in this thesis is based on the knowledge graph,which can be used to connect the entities in the real scene through relationships as the source of side information.Ripplenet model used in this thesis is an end-to-end architecture,which effectively alleviates the limitations of existing knowledge graph recommendation algorithms based on embedding and path.This paper focuses on film recommendation and predict the click through rate of users based on Ripplenet model,which considers the users’ historical interest spreading out layer by layer along the links in the knowledge graph.First of all,the existing common knowledge graph construction technologies are investigated and summarized.Then,a new construction method which is more suitable for movie recommendation for small dataset are proposed.Meanwhile,various relationship types between entities are taken in consideration.The original date of knowledge graph is processed to fit the following recommendation algorithm.After that,Ripplenet algorithm is carried out on the knowledge graph which is built up based on the proposed construction method.Users’ preferences spread over knowledge graph and output the predict users’ click rate.Finally,the solution of cold start problem based on knowledge graph is discussed.The results show that:(1)the knowledge graph construction method proposed in this paper is more suitable for small dataset movie recommendation,and the accuracy is improved by 1.5 percent;(2)the introduction of more relationship types between entities helps to find out the attributes that users prefer when choosing movies;(3)knowledge graph has certain advantages in solving sparse matrix and cold start problems.