Inferring social ties between users with human location history

The location-based social networks have been becoming flourishing in recent years. In this paper, we aim to estimate the similarity between users according to their physical location histories (represented by GPS trajectories). This similarity can be regarded as a potential social tie between users, thereby enabling friend and location recommendations. Different from previous work using social structures or directly matching users' physical locations, this approach model a user's GPS trajectories with a semantic location history (SLH), e.g., shopping malls -> restaurants -> cinemas. Then, we measure the similarity between different users' SLHs by using our maximal travel match (MTM) algorithm. The advantage of our approach lies in two aspects. First, SLH carries more semantic meanings of a user's interests beyond low-level geographic positions. Second, our approach can estimate the similarity between two users without overlaps in the geographic spaces, e.g., people living in different cities. When matching SLHs, we consider the sequential property, the granularity and the popularity of semantic locations. We evaluate our method based on a real-world GPS dataset collected by 109 users in a period of 1 year. The results show that SLH outperforms a physicallocation-based approach and MTM is more effective than several widely used sequence matching approaches given this application scenario.