TSBM: The Temporal-Spatial Bayesian Model for Location Prediction in Social Networks

Abstract

In social networks, predicting a user's locations through those of his or her friends mainly relies on the selection method of the most influential friends of the user, which most of the existing location prediction methods fail to attach importance to. In this paper, we firstly present an analytical procedure in regard to the calculation of the theoretical maximum accuracy for location prediction by virtue of friends' locations. We further compare the theoretical maximum accuracy with the accuracy obtained by the current state-of-the-art methods, and propose an influential friend selection strategy, hoping to narrow the gap between them. More precisely, we define several features to measure the friends' influence on a user's locations, based on which we put forth a sequential random walk with restart procedure to rank the friends in terms of their influence. By dynamically selecting the top N influential friends of the user per time slice, we propose a temporal-spatial Bayesian model to characterize the dynamics of friends' influence for location prediction. Experiments on real data sets prove the effectiveness of our location prediction framework.