Mobility Pattern Based Relationship Inference from Spatiotemporal Data.

The popularity of location-based services and the ubiquity of Internet of Things (ToT) devices have resulted in rich spatiotemporal data. These data enable researchers to study people's social relationship based on their co-occurrences and many inference models were proposed. However, there are still two challenges: How to distinguish co-occurrences between acquaintances and strangers? What kind of co-occurrence contributes to strong social strength? In this paper, we propose a mobility pattern based relationship inference model (MPRI) to address above challenges. We extract mobility patterns from spatiotemporal data and adopt them to characterize co-occurrences. A classification model is trained for social relationship inference. The experimental results on two real-world datasets demonstrate that the proposed MPRI model can properly differentiate co-occurrences by simultaneously considering spatial and temporal features. The comparison results also indicate that MPRI model significantly outperforms state-of-the-art social relationship inference models.