| With the rapid development of film and television industry,users are in the trouble of information overload and misdirection.Personalized recommendation system is an effective way to solve the problem,but the existing system has the problem of data sparsity.At present,there are many methods to solve data sparsity,and the most representative one that can be used on a large scale is knowledge graph.Under the background of rapid development of deep learning,the recommendation generation based on graph embedding adopts the vectorization method of graph semantic features,effectively combines knowledge graph and recommendation algorithm and gradually develops into a popular one,among which the ripple network framework is the most representative.From the perspective of the improvement and application of the recommendation model,this thesis carries out the research on the knowledge graph-assisted movie recommendation system in the field of film.The main work is as follows:(1)Construct the movie knowledge graph.Based on the analysis of domain knowledge,the relevant knowledge elements are abstracted,and the ontology library is constructed based on Protege.The relevant knowledge is crawled from the Douban film website and stored in My SQL,and knowledge extraction is carried out based on the My SQL database and D2 RQ platform.Finally,considering the characteristics of Neo4 j,in order to better retrieve knowledge,it is transferred to Neo4 j for persistent storage,and the knowledge graph of movie domain is constructed,making full preparation for the subsequent downstream task recommendation system.(2)A Combining Knowledge graph for Preference propagation recommendation model(CKP)is constructed.Based on the assumption that a user’s preference for a project mostly depends on the attributes of the project,this thesis deduces the user preference by combining the project knowledge graph and user-item interaction matrix,and then introduces the user preference into the preference propagation model to predict user click rate.This preference propagation model is improved based on the Ripple Net framework,and relevant experiments are carried out based on two public data sets of film and music.AUC,ACC and F1 are taken as the evaluation indexes,and the effectiveness of the knowledge graph introduction and the improved scheme is verified through the analysis of hyper-parameters sensitivity,comparison of the results of the baseline models on CTR,data sparsity experiment and ablation experiment.Compared with the original model,the AUC index of CKP model was increased by 1.4% and the ACC index was increased by 2.4%.In the music data set,AUC index was increased by 6.3% and ACC index was increased by 9.6%,which highlighted the superiority of the improved model.(3)Design and implement a personalized movie recommendation system.The system takes reducing users’ difficulty in finding information and mining users’ potential interest items as two targeted goals.The system provides movie search,entity query,graph question and answer to help users improve the efficiency of searching.At the same time,the interactive data in the system is stored in the My SQL database,and combined with the movie knowledge graph stored in Neo4 j,the established recommendation model is introduced to match the appropriate movie for the user,so as to improve the user experience and satisfaction.On the premise of clear requirements,the system functions and database are designed in detail,and related functions of the client side and the management side are developed and realized.Through testing,the stability and accuracy of the system are ensured. |
