Design And Implementation Of Recommendation System Based On Muiti-modal Knowledge Graph

In recent years,with the rapid development of the Internet,more and more text,pictures,videos and other content have been published on the Internet.The explosive growth of Internet information poses a new challenge to computer technology.The early development of the Internet,the demand of information to the people is simple,usually can clearly describe the problem,through search engines to find the answer,and now,some of the basic information needs can be met already.For how to further explore users potential demand,and help users to find out more that may be of interest to the content,personalized recommendation system solution are given.At present,a common solution is to build a hybrid recommendation system,based on collaborative filtering and external auxiliary information,to solve the problems of data sparsity and cold start.As a kind of very useful auxiliary information,knowledge graph can effectively improve the recommendation effect by combining with collaborative filtering recommendation algorithm.In recent years,with the development and prosperity of short video,the recommendation system has put forward more business requirements related to multi-modal information.Therefore,this paper focuses on the recommendation system based on multi-modal knowledge graph.The main contents of this paper are as follows:(1)Collaborative multi-modal knowledge map is constructed.Collaborative multi-modal knowledge graph is constructed by fusing multi-modal information and user-item interaction data into the knowledge graph.In the multi-modal information processing,the pre-training model is used to extract feature vectors.In addition,keywords extraction technology is innovatively used to extract keywords from multi-modal information,so as to enrich the connection relationship between items and achieve better recommendation effect.(2)The multi-modal knowledge graph attention network model MM-KGAT is proposed.A MM-KGAT model with better recommendation effect was proposed by improving each layer of KGAT model and integrating multi-modal entities.The representation learning method is optimized in the feature coding layer,and a new and better aggregation method is proposed in the information aggregation layer to improve the recommendation effect.(3)A movie recommendation system is set up.Through the front and back end interaction technology,a movie recommendation website is constructed,which provides a complete solution for the algorithm landing.The system provides users with rich functions and visual operation interface,so as to facilitate users to experience the recommendation effect of algorithm.