A Hybrid Method for POI Recommendation: Combining Check-In Count, Geographical Information and Reviews

Abstract

Due to the rapid development of mobile devices, global position systems (GPS), Web 2.0 and location-based social networks (LBSNs) have attracted millions of users to share their locations or experiences. Point of Interest (POI) recommendation plays an important role in exploring attractive locations. POI recommendation is associated with multi-dimensional factors, such as check-in counts, geographical influence, and review text. Although GeoMF can model geographical influence well by matrix factorization (MF), it ignores the impact of review text for POI recommendation. We propose a hybrid method to joint check-in counts, geographical information and reviews for POI recommendation. Specifically, we connect check-in counts and geographical information by incorporating geographical information into matrix factorization. In addition, we combine check-in counts with review text by aligning latent check-in counts in MF, and utilize hidden review topics obtained from LDA by a transformation. The results of our experiments on the real-world dataset show that our proposed model can improve the performance of recommendation.