A POI-Sensitive Knowledge Graph Based Service Recommendation Method

In the past years, "group psychology" has been fully utilized in service recommendation. The wide use of collaborative filtering recommendation and user's trust in the rating of service have proved its success. Content-based recommendation further optimizes the recommendation accuracy by strengthening the analysis on service contents. However, existing works pay little attention to the content details and their underlying semantic correlations. This paper proposes a knowledge graph based method that comprehensively considers both community effect and details of service contents. We build a knowledge graph for representing service supplydemand networks, and incorporate the community network structures of users and services into it. Further, we give a method of mining POIs (Point of Interests) of users from user reviews based on RAKE text mining algorithm and incorporate POIs into the knowledge graph. User reviews, no matter praise or criticism, always imply user preferences on specific service contents. Finally, we design a recommendation algorithm POIKG RS based on Representation Learning method of Knowledge Graph. A set of experiments demonstrate that the proposed approach can better utilize the characteristics of different group of users and their POIs, resulting in improved recommendation accuracy.