Recommendation System Based On Knowledge Graph Attention Network

In the context of big data in recent years,recommender systems usually use collaborative filtering and deep learning techniques to mine effective information from massive data to improve recommendation performance.However,collaborative filtering technology only uses useritem interaction data for recommendation without auxiliary information,so it cannot deeply mine the connection between users and items,and faces data sparsity and cold-start problems.At the same time,most of the recommendation models based on deep learning are not interpretable,resulting in the recommendation system unable to provide reasonable recommendation explanations and poor user experience.In order to alleviate these problems,the current mainstream solution is to introduce auxiliary information into the recommender system.As a semantic knowledge base,knowledge graph can contain rich auxiliary information,and the recommendation system based on knowledge graph has gradually become a research hotspot.However,the current recommendation system based on knowledge graph still faces many problems in the research and application:(1)the knowledge graph constructed for the recommendation dataset has interference information;(2)the recommendation model fails to effectively combine the semantic information of the knowledge graph with the collaborative information contained in user-item interaction data;(3)the recommendation system cannot provide recommendation explanations,and the recommendation model itself is not interpretable;(4)the deep learning model used for recommendation is complex in structure and takes too long to train,so it is not suitable for practical application scenarios.In view of the above problems,this paper proposes and implements a recommendation system based on knowledge graph attention network,which uses knowledge graph to enhance the connection between users and items to achieve more accurate recommendation service,and combines attention network and recommendation explanation generation algorithm to make recommendation The system is interpretable.The main research work of the paper includes:(1)Aiming at the problems that the traditional knowledge graph construction method has too much workload and there is interference information in the constructed knowledge graph,a connection-enhanced knowledge graph construction method is proposed,which utilizes the link between the recommendation dataset and the open domain knowledge base.Introducing auxiliary information from open domain knowledge bases into recommender systems to enhance the connection between users and items.(2)Aiming at the problems that the recommendation model based on deep learning lacks interpretability,takes too long to train,and does not combine semantic information and collaborative information,a Collaborative Multi-View Attention Network for recommendation prediction(Collaborative Multi-View Attention Network,CMVAN)recommendation model,which integrates the effective auxiliary information in the knowledge graph from the perspectives of users and items,and realizes the effective combination of collaborative information and semantic information,so as to deeply mine the connection between users and items to improve the recommendation performance of the model.Through comparative experiments on public datasets in three recommendation domains,the effectiveness of the CMVAN model in terms of accuracy and training cost is verified.(3)Aiming at the problem that the recommendation system does not have the inherent interpretability of the model,or cannot provide recommendation explanations,this paper proposes a recommendation explanation generation algorithm based on the CMVAN recommendation model,which combines the relationship between the attention network weight and the knowledge graph,locate the key entities that the user is interested in,generate the recommended explanation in the form of text,and realize the interpretability of the recommender system.(4)Based on the above recommendation model and explanation generation algorithm,an explainable recommendation system based on knowledge graph is designed and implemented to verify the effectiveness of the recommendation model and explanation generation algorithm proposed in this paper in practical application scenarios.