Mining User Similarity Based on Users Trajectories

In this paper, we first extract latent topics of users´ check in by topic model. In order to learn the geographic features of topics, we analyze the distribution and correlation of the venues of each topic. Based on the assumption that each latent topic, to some extent, reflects some users´ habits or behavior patterns in certain area, we use the latent topics to mining user similarity. When use the latent topics to mining user similarity, we first calculate the entropy of each topic, and then use it to analyze which topics will be salient when calculating user similarity. And we define two sets of topic. One is common topic set. The other is variability topic set. Subsequently, we propose two models to calculate user similarity. One is single-layer model. The other is two-layer model. Two-layer model is improved to single model. Finally, we use the proposed models to calculate user similarity in all topic set, common topic set and variability topic set. And then we verify our models by experiment.