Research On Personalized Recommendation Algorithm Based On Knowledge Graph And Evidence Theory

Knowledge Graph is defined as a multi-relationship graph containing multi-types of nodes and multi-types of edges.It is actually a knowledge base of semantic network,and has been widely used in securities investment,search,adaptive education,big data risk control,chatbot,intelligent medical care,recommendation system and other fields.How to take effective auxiliary information to improve the performance of collaborative filtering recommendation system has become a research hotspot.At present,the main method is introduced in the recommendation system to improve,such as deep learning,neural network,knowledge map and other auxiliary information,compared with other auxiliary information,and knowledge map not only can make with richer semantic relations between entities,to be able to dig more deeply the user's interest,and can realize different kinds of entity relationship connection,Therefore,in the application scenarios such as film recommendation and music recommendation,introducing knowledge graph and evidence theory into recommendation algorithm can achieve good results.The main research contents of this paper are as follows:(1)on the research on the technology of knowledge map building,knowledge extraction technology is not only the first step to build,and is the key of the knowledge map of quality assurance,thus this article focuses on the knowledge of related technologies,including the named entity recognition and the relationship between extraction and so on,through comprehensive analysis,comprehensive performance is selected optimal combination model for knowledge extraction,Thus,the task of constructing the knowledge graph is realized,which lays a foundation for the follow-up research.(2)in view of the collaborative filtering recommendation system because of its scarcity and almost no user behavior data in the data and the problem that the lead to recommend the effect not beautiful,in this paper,through the research,put forward a new method based on knowledge and information fusion,specifically,first of all need to build technical film knowledge map construction,and then in the user's social network community,and then,Method was developed to study knowledge in every community transmission of preferences and modeling the flexibility of the information fusion theory,and put them together,they have their own advantages,to achieve a score of users and commodities matrix improvement purpose,finally can interact through a few existing users and items information,mutual information of users and commodities for which the forecast.(3)Carry out experiments and analysis according to the previous relevant studies.Firstly,due to the quality differences among different construction technologies,the effects of different knowledge extraction capabilities and the quality of knowledge graph construction on recommendation performance were verified in the experiment.Then,the influence of the introduction of knowledge graph on recommendation performance is verified.Finally,the relevant model was selected as the baseline,and the parameters consistent with the baseline were set for the experiment.The feasibility and effectiveness of the improved model and method were verified by comparative analysis.