Research On Personalized Recommendation System Based On Knowledge Graph

Nowadays,with the explosion of Internet information,how to provide users with personalized recommendations and achieve efficient transmission of information has become an urgent problem in the current recommendation system research.In the recommendation,it is logically that integrated into the auxiliary information such as social network,comment data and hotspot can effectively improve the performance of the recommendation model.In recent years,the research on the combination of knowledge graph and recommendation system has attracted widespread attention.By studying the potential connection of entities between the recommendation scene and knowledge graph,it can provide rich complementary information for item recommendation.However,most of the existing studies have not effectively established the information interaction between the recommendation model and the knowledge graph,which limits the extraction effect of knowledge information,and is also susceptible to interference from irrelevant information.Therefore,this paper studies the efficient combination of recommendation system and knowledge graph.This paper studies the strategy of combining knowledge graph with recommender system in different knowledge density scenarios.Firstly,this paper proposes a recommendation scheme based on graph convolutional network in knowledge-intensive scenarios.In order to effectively integrate the knowledge graph information into the recommendation model,we designed a neighborhood aggregation collaborative filtering algorithm combining the knowledge graph.The model uses knowledge graph to expand and extract users' potential interests,and then iteratively embeds features with attention bias into users' characteristics.Secondly,this paper proposes a multi-tasking learning scheme to complete the recommended tasks in knowledge sparse scenarios.On the basis of the above research,this paper further explores the effective recommendation scheme for knowledge sparse scenarios,and proposes an alternate training algorithm for recommendation and knowledge graph embedding.It consists of recommendation module and knowledge embedding module,which extracts the potential features of users and items by using the information of knowledge triples and transfers the knowledge features to the recommendation module in the process of alternating training.Finally,this paper completed a series of experiments on five public datasets and verifies the effectiveness of the above recommended scheme in the corresponding scenarios.The models proposed in this paper are ahead of the latest models in terms of click-through rate prediction and Top-K recommendation.