Abstract: The purpose of recommender systems is to solve the problem of “information overload”. However, most of the current research methods use user and commodity information to model user interest, and do not consider the use of knowledge map to build user interest and image at the same time to enhance the performance of recommender systems. Therefore, this paper proposes a personalized recommender model integrating knowledge map and comment text. Firstly, different types of know- ledge maps are used to associate user items and user comment texts to expand user interest and extract entities from comment texts. Secondly, interest features with user interest preference are obtained by constructing user interest network. Then, by constructing the portrait network of the portrait module and the emotion module, the portrait features with the users emotion tendency are extracted. Finally, the decision-making layer is used to predict the click through rate. The experimental comparison and analysis were carried out on Amazon datasets. Firstly, the recommendation performance of the proposed model is evaluated, and then is compared with the current recommendation model integrating knowledge map and comment text, which verifies the effectiveness of the proposed model.

Key words: recommended system, knowledge graph, weighted heterogeneous graph, interest attention