Location-Based Recommendation Using Incremental Tensor Factorization Model

Newly emerging location-based online social services, such as Meetup and Douban, have experienced increased popularity and rapid growth. The classical Matrix Factorization methods usually only consider the user-item matrix. Recently, Researchers have extended the matrix adding location context as a tensor and used the Tensor Factorization methods for this scenario. However, in real scenario, the users and events are changing over time, the classical Tensor Factorization methods suffers the limitation that it can only be applied for static settings. In this paper, we propose a general Incremental Tensor Factorization model, which models the appearance changes of a tensor by adaptively updating its previous factorized components rather than recomputing them on the whole data every time the data changed. Experiments show that the proposed methods can offer more effective recommendations than baselines, and significantly improve the efficiency of providing location recommendations.