Research On Movie Recommendation Technology Based On Knowledge Graph

With decades of development in the age of data and information,the volume of data on the Internet has also become larger and larger.When people searched,they encountered a large amount of data information in different data formats,which caused information overload.Among them,the rapid growth of multi-dimensional data information in the field of movies brings massive resources to users while also preventing users from effectively finding movies they are interested.The movie recommendation system can efficiently search for favorite movies for users by filtering redundant information.In recent years,this system has made it easy for users to search for movies.However,the performance of traditional recommendation systems is still limited by data sparsity and cold start.The main reason is that each user has different interests and hobbies,so the preference for movies also varies from person to person.Since there is little comment data in the system,user data cannot be obtained,so the user's preferences cannot be correctly predicted,which affects the recommendation effect.In order to mitigate the impact of data sparsity,building a knowledge graph in the field of movies,using the knowledge graph can calculate the semantic similarity between movies,and combines traditional recommendation algorithms to achieve personalized movie recommendations.The research content is as follows:Firstly,the design and implementation of knowledge graph in the field of film are introduced.The translation model has the advantages of lower complexity and higher accuracy in the process of constructing the knowledge graph,so using a translation model with less calculation to construct the knowledge graph.The traditional translation model Trans E cannot deal with the multiple relationships between movie entities,which leads to a low accuracy in the calculation of similarity in the recommendation algorithm.Therefore,using an improved Trans HR model to make up for the shortcomings of the Trans E model in the representation of multiple relationships and improve the recommendation performance.Secondly,in order to further improve the accuracy of the knowledge graph in calculating the similarity of movies,the clustering algorithm is used to mine the same attributes of movies in movie reviews and embed the knowledge graph as a category relationship in the process of constructing the knowledge graph in the movie domain.Combined with the K-means clustering algorithm to extract hidden relationships between entities in movie reviews and add them to the knowledge graph to improve the integrity of the knowledge graph and further improve the recommendation performance.Finally,the similarity between movies is calculated by combining knowledge graphs and the similarity of movies based on user rating matrix is applied to matrix decomposition to form a personalized recommendation algorithm.This method makes use of the existing knowledge base to make up for the problem of decreased recommendation performance caused by less data in the recommendation process.The comparison experiment results show that the algorithm improves the recommended recall rate,accuracy,RSME and MAE and other indicators compared with the traditional algorithm.