| As an important information filtering and personalized service technology,recommendation system is widely used in various fields.Although the traditional recommendation algorithm model has been widely used,the recommendation performance of the traditional recommendation algorithm is not high.The knowledge graph contains a wealth of entities and attributes.By analyzing the relationship between entities in the knowledge graph and mining potential correlations and similarities to achieve recommendations,it can well solve the problem of data sparsity and cold start.The Ripple Net recommendation model based on knowledge graphs introduces vector representations and makes full use of entity connection relationships to mine high-order semantics,enabling accurate recommendations.However,the Ripple Net recommendation model does not fully consider the importance of data.In order to further improve the accuracy of recommendation,the main research contents of this paper are as follows:(1)This paper proposes a method to eliminate data redundancy and improve data relevance by constructing a maximal subnetwork.First,the proposed subnetwork extraction algorithm is used to extract the largest subnetwork whose nodes and edges can well reflect the characteristics of the entire network.Then,the original data sets Book-Crossing and Movie Lens-1M were compared with the extracted largest subnets Book＿Max Sub Net and Movie＿Max Sub Net on the Ripple Net model.In the CTR click-through rate prediction scenario,the experimental results show that for the Book-Crossing dataset,the AUC value increases from 72.2% to 72.7%;for the Movie Lens-1M dataset,the AUC value increases from 91.3% to 91.9%.Experiments demonstrate that the recommendation accuracy can be improved by extracting the largest subnetwork.(2)This paper proposes a hybrid recommendation model(HRCF),which fuses the Ripple Net model with the recommendation results of a user-based collaborative filtering algorithm.Specifically,the model improves the effectiveness of the recommendation model by performing weighted fusion of the results of two recommendation models.Finally,in the click-through rate(CTR)prediction scenario,using the Book-Crossing dataset as the experimental data,the HRCF model was compared with the Ripple Net model,and the results showed that the AUC value increased from 72.2% to 73.1%.At the same time,on the Movie Lens-1M dataset,the AUC value also increased from 91.3% to 92.1%.(3)Finally,this paper designs and builds a movie recommendation system based on two improved recommendation models.Firstly,the overall structure of the system is designed,and then each functional module of the system is realized.Through the specific application,the effectiveness of the above method is proved. |
