Deep Matrix Factorization Recommendation Model Based On Knowledge Graph And User Preferences

As an important technology to filter mass information,recommendation system has been successfully applied in various practical scenarios.The most widely used algorithm in the current recommendation system is the collaborative filtering algorithm and the deep neural network which has developed rapidly in recent years.However,the former needs to face the problems of cold startup and sparse data,while the latter can improve the performance of the recommendation,but has very weak interpretability.Therefore,people begin to pay attention to the knowledge graph with rich semantic information,which can not only improve the recommendation performance,but also enhance the interpretability of the system.In this paper,a deep matrix factorization model(KPDMF)based on knowledge graph and user preferences is proposed.Based on the assumption that user likes item may be due to certain attributes of the item,this paper uses neural correlation model and attention mechanism to learn user preferences,and then introduces user preferences into the deep matrix factorization model,so that the model can learn more effective implicit vectors.On the other hand,because there is a certain correspondence between the items in the recommendation model and the entities in the knowledge graph,this paper uses a multi-task learning framework to combine the knowledge graph with the deep recommendation module based on user preferences,and cross-transfers the information of the items and entities through the cross-compression unit to achieve information complementarity.In addition,considering that the relations in KG are related to item attributes,the relations and user preferences are cross-studied to better explain the user preferencesIn addition,we further compared the above KPDMF model with the models which have better performance in the same field in recent years on several evaluation indexes in the three datasets of movie,book and music.The experimental results show that both on the recommended performance and the performance in knowledge graph,KPDMF model's performance is obviously superior to other models,and in the case of sparse data,the KPDMF model's performance is still very prominent.