Personalized Point-of-Interest Recommendation with Relation-Enhanced Graph Convolutional Network

Abstract

Point-of-Interest (POI) recommendation recommends different personalized services to interested users, which are widely used in people's daily life. However, with the massive increase in users and POIs, the POI recommendation system faces the following challenging problems: (1) The results of the recommendation service are not personalized enough, and little attention is paid to the details and semantic relevance. (2) Often face the problem of cold start (3) Recommended accuracy can be further improved. In order to cope with these difficulties, this paper proposes a POI recommendation method POI-Graph Convolutional Network with relation-enhanced Graph Convolutional Network(P-GCN). First, this paper mining the user's preferred POI information with a quantitative emotion from the user's comment information and historical information in the Yelp and DianPing datasets, and builds a representation of the relationship between the service user preferences knowledge graph (KG). On this basis, the P-GCN model is introduced, and uses end-to-end learning to mine the associated attributes between projects, and to make the high-order structural information and semantic information in the knowledge graph discovered. Finally, through the method of node aggregation of multi-hop neighbor node information, the high-order semantic relevance of user nodes and their potential preferences are obtained, and POI recommendations are made to users. Through extensive experiments, our method can better utilize the characteristics and POI of different users who own the same preferences, which can improve the precision and personalization of recommendation. The experimental results are superior to the baseline method in terms of evaluation indexes.