A List-wise Matrix Factorization Based POI Recommendation by Fusing Multi-Tag, Social and Geographical Influences

Abstract

In this paper, a list-wise matrix factorization method is proposed for point of interest (POI) recommendation that fuses multi-tag, social and geographical influences. Based on the relevance of location information in the physical world, location-based networks (LBSN) contain a new social structure. Firstly, we extract and model mutil-tag, social and geographical influences separately from three layers of LBSN. For multi-tag influence, we extract a user-tag matrix from the initial user-POI rating matrix by analyzing the relations between POI and the related bag of tags. For social influence, we model the social influences by using social regularization method and considering distance factor between trusted users. For geographical influence, an effective method to model the geographical influence is proposed by considering the location of user and POI and the related region center. Secondly, in order to improve the performance of point of interest (POI) recommendation, we include multi-tag influences and fuse the social information and geographical influences into a list-wise matrix factorization (MF) framework for making prediction of recommendation list. The experimental evaluation is conducted on Yelp datasets with different scales. Our experiments show the proposed method significantly outperforms other state-of-the- art recommendation approaches and achieve a great result for POI recommendation