Research On Personalized Recommendation Method Based On Knowledge Graph Fusing Users' Explicit And Implicit Preferences

With the high-speed development of science and technology,although the era of big data brings convenience to people,it is also accompanied the problem of data redundancy.In the face of massive data,users cannot make a correct and appropriate choice.Search engines can help people solve a small part of the data redundancy problem,but it cannot provide personalized recommendations for users.Therefore,the recommendation system application is born.Personalized recommendation system provides users with the most appropriate data and information by screening out useless data.However,the traditional recommendation system still encounters the problem of data sparsity and cold start.Therefore,a lot of researchers hope to improve the recommendation effect through auxiliary information.For example,combining knowledge graph with recommendation system and fusing the information into the recommender system.In this thesis,knowledge graph is introduced into the recommendation system to describe users' implicit attribute preferences by fusing the network edge structure information of knowledge graph,we implemented to integrate users' explicit and implicit preferences.The problem of sparse data is solved,the performance of recommendation is improved,and the interpretability of recommendation system is enhanced.Finally,a personalized recommendation system is developed.The main research contents and innovations of this thesis are as follows:1.This thesis uses the network edge structure information of knowledge graph to mine users' implicit attribute preference.Starting from the user's historical record,the embedded of user's id and the project is as the original input.The item embedding represents that each item I is associated with item embedding in the vector space,by comparing item I with the header entity hi and the relational entity r_i to calculate implicit preference inheritance probability for each triple.By multiplying the implicit correlation probability and explicit correlation probability with t_i to expressed propagation vector of the next hop.Thorough recursive calculation to get the final user embedded.Finally,the click-through rate of the project is predicted through user embedding and project embedding.2.This thesis developed a personalized movie recommendation system.We Analyze the requirements of the system and determine the overall framework of the recommendation system.Then combined with the actual scenario design to realize the functional sections,and finally tested the system to verify whether the system meets our expected requirements.