Research On Movie Recommendation System Based On Knowledge Graph

With the rapid development of Internet technology,people can obtain a large amount of data and information through the network,but they are also troubled by the choice caused by information overload.In order to alleviate the burden of massive data information,the recommendation system came into being.As an information filtering method,its purpose is to provide users with satisfactory personalized products and services.With its own advantages,the recommendation system has been widely used in e-commerce,social leisure,intelligent life and other fields.However,when the system is faced with new users joining or users only generating very few interactive behavior situations with the system,due to the lack of sufficient historical information,the recommendation system faces data sparseness and cold start problems,which affects the normal operation of the recommendation system.Introducing knowledge graph into recommendation system,using the rich semantic information provided by knowledge graph can effectively alleviate the above problems and improve the accuracy,diversity and interpretability of recommendation results.In this paper,the knowledge graph of the film field is constructed for film recommendation.In order to further improve the recommendation performance,the structure and result presentation of the existing knowledge graph are improved.At the same time,the attention mechanism and graph convolutional neural network are integrated.A new film recommendation algorithm based on knowledge graph is proposed.The algorithm can effectively alleviate the data sparseness and cold start problems in the recommendation system and achieve good recommendation results.The main research contents of the article are as follows :(1)The existing recommendation algorithms based on knowledge graph mostly query movies according to users ’ preferences when constructing movie knowledge graph.However,when recommending related movies,due to the complexity of the connection between movie entities,the difficulty of constructing knowledge graph will increase accordingly,which will lead to the decrease of query efficiency,make the visualization effect of knowledge graph worse,and affect the user ’s experience.In view of the above problems,this paper proposes a new method to construct the knowledge graph of movies.Firstly,by calculating the similarity between movie entities,the unnecessary connection relationship is reduced and the quality of knowledge graph is improved.Secondly,the redundant paths in the knowledge graph are deleted by path hiding technology to reduce the complexity of the knowledge graph,thereby improving the query efficiency of the knowledge graph,making the visualization effect clearer and easier to understand.Correspondingly,users can browse and search the knowledge graph more conveniently and improve the user experience.(2)When the existing knowledge graph-based recommendation algorithm performs model training,if the newly added users lack sufficient feature information or historical interaction behavior data,it will make the recommendation algorithm difficult to capture the user ’s preferences,resulting in the model unable to recommend its suitable movie list to the user when making recommendation decisions.Aiming at the above problems,this paper proposes an Adaptive Knowledge Graph Convolutional Networks(AKGCN)recommendation algorithm.The algorithm uses the graph convolutional neural network as the model framework to learn the node representation vector in the knowledge graph,and passes the feature information of its adjacent nodes to each node through the graph convolution operation.In addition,when aggregating the feature vectors of users and movies,the model adaptively extracts the relevant feature information according to the attention mechanism,thereby improving the generalization ability and recommendation accuracy of the algorithm.The AKGCN algorithm proposed in this paper effectively solves the problem of data sparseness and cold start in the recommendation system.(3)In this paper,the Movielens-1M movie dataset collected by the University of Minnesota in the United States is selected for experiments.This dataset is one of the most widely used and classic datasets in the field of recommendation systems.In addition,the performance of each algorithm is evaluated based on commonly used evaluation indicators in recommendation systems such as precision,recall,and normalized discounted cumulative gain.The experimental results show that the AKGCN algorithm proposed in this paper performs better than other recommendation algorithms when dealing with the sparse user interaction behavior or the cold start problem of new users in the system.