Multi-Context-Aware Location Recommendation Using Tensor Decomposition

Abstract

With the rapid growth of information generated by online social network platforms and the increased usage of Location-Based Social Networks, location recommendation research has attracted more attention both in academic and industry. However, the problem of data sparsity still posses a severe challenge to the existing location recommendation methods. Moreover, extracting and modeling multiple contextual information, which is one of the key factors that influences user check-in preferences, is another big challenge faced by the existing methods. Many of the existing location recommendation methods have low accuracy because they utilize limited contextual information when modeling user check-in behaviors. In this paper, we propose a Multi-Context-aware Location Recommendation using Tensor Decomposition (MCLR-TD) approach that incorporates multiple context information at different granularity scales in modeling user check-in behavior. We use a four mode tensor to model the relationship among the four dimensions: users, locations, time and weather. In order to reduce the data sparsity problem, we further construct four feature matrices that are collaboratively decomposed with the tensor. We carry out extensive experiments on two real-world datasets collected from Foursquare and Yelp and the results demonstrate the effectiveness of our approach.