Fused matrix factorization with multi-tag, social and geographical influences for POI recommendation

Abstract

We make plans to provid a point-of-interests (POI) recommendation method for location-based social networks (LBSNs) in the paper. LBSNs contain a special three layers network structure based on relevance of location information in the physical world. The available information of three layers of LBSN makes it possible to mine the internal correlations between different layers and to analze users’ preference on locations. We try to model the muti-tag, social and geographical influences separately form three layers of LBSN. We first model the mutli-tag influences via extracting a user-tag matrix from intial uer-POI matrix. Next, we introduce social regularization to model the social influences, Thirdly, we use a normalized function to model the geographical influences. Accordingly, we include multi-tag influences and fuse the social regularization, geographical influence into a famous matrix factorization(MF) framework. Finally, we conduct extensive performance evaluations on the large-scale Yelp datasets. As a result, the fusion framework outperforms other state-of-the-art approaches on recommendation.