Research On Improved RippleNet Recommendation Method

The recommender system can help users discover their potential interested items in the era of information explosion,and it also brings benefits to companies,therefore the research of recommendation system is of great value and significance.The collaborative filtering recommendation algorithm based on the user item interaction information has been researched for many years,and its recommendation quality is hardly improved significantly.With the rapid development of knowledge graph theory and graph neural network technology in recent years,knowledge graph-assisted recommendation system has become a new research trend to further improve recommendation quality.RippleNet is a typical recommendation method based on knowledge graph with good recommendation quality,but it still has the problem of insufficient information utilization in feature extraction.Therefore,based on RippleNet,through the fusion of the two models of RippleNet,the user features and item features are jointly extracted on the knowledge graph,thereby constructing a new recommendation model,which can better discover user interests and the distribution of item features on the knowledge graph and the relationship between them further digs out the user's interest in the item,thereby improving the recommendation quality.According to the constructed recommendation model based on knowledge graph,a prototype recommender system was designed and implemented.Representative public data sets were selected in the three fields of film,music and book,and several typical recommendation methods were selected to compare with the proposed recommendation model.Through comparison experiments,it is found that the proposed recommendation method is superior to the compared recommendation methods in the recommender system evaluation index,which verifies the effectiveness of the recommended method.