Research On Personalized Recommendation System Based On Knowledge Graph

In the era of big data,Internet technology has brought people massive data and greatly enriching people’s lives,but at the same time,it has become very difficult for people to screen out the content they really need.The advent and application of personalized recommendation systems has greatly improved this situation and improve the user experience.Among the technologies of personalized recommendation systems,collaborative filtering algorithms are mature and widely used.However,such traditional recommendation algorithms often suffer from data sparseness and cold start.In order to alleviate these problems,using the side information to improve recommender systems has become a research hotspot.Among them,knowledge graph has a significant effect on recommender systems with its rich structured information and semantic information.At the same time,the development of graph neural network technology has made it useful in the research of recommender systems.In this context,this paper focuses on the research of personalized recommendation system based on knowledge graph,and analyzes the advantages and disadvantages of existing technologies.Combined with the current cutting-edge research trends,the existing problems are targeted research and experiments.The main contributions of this paper are:(1)Aiming at the problem of insufficient use of high-order information of knowledge graph by recommender system,we propose a Knowledge-aware Recommendation Algorithm based on Hierarchical Attention Network(KGHAN)based on hierarchical attention mechanism.Combine the user-item graph and the knowledge graph into a collaborative knowledge graph,and carry out the dissemination of knowledge perception on this unified graph.When collecting multi-hop neighborhood nodes of an entity,the attention weights are considered from the levels of nodes and relationships in a way based on the hierarchical attention mechanism,and the attention scores of the two levels are fused.The multi-layer attention embedding layers are then aggregated to obtain the final representation vector of users and items.It effectively aggregates the higher-order semantic information and structural information in the knowledge graph.Experiments on two public datasets show that the proposed method effectively improves the accuracy and diversity of recommendations.(2)Aiming at the problem of insufficient use of collaborative information in the process of knowledge graph assisted recommendation,we propose a knowledge graph attentive network combined with collaborative information for recommendation(KGAT-CI),which adopts an end-to-end framework.The model combines the collaborative information in the user-item interaction record and the side information of the knowledge graph,then it uses the graph attention network to learn the embedding of the neighborhood.Finally,the model obtains the representation of users and items for recommendation prediction.Experiments were carried out on datasets in three different domains,and the results showed that the model performance was improved.(3)Aiming at the problem of lack of fine-grained consideration of collaborative information in knowledge graph recommendation,based on the attention network method of knowledge perception,we propose a knowledge graph attention network combined with the user’s perspective for recommendation(UPKAN).The model further refines the collaborative information representing users by modeling perspective factors,and uses the aggregation of relational paths in the perspective module to retain the relevant structure and semantic information in the knowledge graph.Combined with the knowledge graph attention network on the item side,the attention mechanism is used to embedding and aggregate the knowledge perception of the neighborhood to further enrich the user and item representation.Experiments on two public datasets show that the model performs significantly better than other baseline methods.