Research On Recommendation Method Based On Knowledge Graph And Enhanced Features

With the increasing development of Internet technology,the rapid growth of data in the Internet,human beings are also plagued by the problem of information overload,so researchers propose a recommendation system to make personalized recommendations for users and reduce the impact of information overload on users.Traditional recommendation systems have the problem of low recommendation accuracy,researchers propose to use some additional information as auxiliary information for recommendation,knowledge graph contains rich semantic information between entities,and its application to the recommendation system can alleviate these problems.Aiming at the problems existing recommendation algorithms based on knowledge graph,this thesis aims to design a recommendation algorithm with higher accuracy and improve user experience.The research content of this thesis is as follows:(1)Users with similar preferences will also like similar items,and the information of these items can be used to capture the potential preferences of users,but most existing recommendation models based on knowledge graphs ignore this.To solve the above problems,this thesis proposes a recommendation model based on high-order collaborative information and important feature mining.Through the historical interaction data of users and objects,the model extracts first-order collaborative information and high-order collaborative information,and uses the attention mechanism to mine important features.Use the user’s first-order neighbor and third-order neighbor in the bipartite diagram of users and items,and the second-order neighbor of the item,corresponding to the entities in the knowledge graph,and perform propagation operations in the knowledge graph to extract the knowledge graph information and expand user preferences.Finally,the two types of information are fused to obtain the final representation of users and items for prediction.Experimental results on two real datasets show that the recommendation accuracy is improved.(2)The above method extracts the characteristics of users and items from knowledge graph and interaction data,which alleviates the problem of data sparseness and cold start.But these features are limited to textual features and do not take into account other types of features.To solve this problem,a two-channel attention mechanism recommendation model based on image features is proposed.Considering the rich information in the item image,the model uses VGG19 to extract the image features related to the item,uses a dual-channel attention mechanism to assign weight to the user’s historical behavior image,and differentially aggregates the user’s historical behavior image.The heterogeneous propagation strategy is adopted to iteratively disseminate in the user-item two-part graph and knowledge graph,and explicitly encode the collaborative information and knowledge graph information.At the same time,when aggregating neighborhood information in the knowledge graph,the attention mechanism is used to assign different weights to the tail entity according to the different head entities and relationships,so as to better extract the semantic information in the knowledge graph.According to the correlation between the user’s historical click item and the current task,it assigns weight to it and dynamically aggregates the user’s historical interaction information.Finally,several types of information are integrated to make recommendations.Multiple sets of experiments show that this method can effectively improve the recommendation accuracy.