From trajectories to activities: a spatio-temporal join approach

People's activity sequences such as eat at a restaurant after 2 hours of shopping, contains rich semantic information. This information can be explored for a broad range of applications and services. However, it is impractical to ask a large number of people to record their daily activities. As the increasing popularity of GPS-enabled mobile devices, a huge amount of trajectories which show people's movement behaviors have been acquiring. The natural link between activities and traveling motivates us to investigate a novel approach to automatically extract sequences of activities from large set of trajectory data. Intuitively, activities can only happen when trajectory is geographically near for a proper period of time for these activities, such as 30 minutes for dining in a restaurant. In this work, the concepts influence and influence duration are proposed to capture the intuition. We also propose two algorithms to join large set of trajectories with activities with duplication reuse techniques. We conduct comprehensive empirical studies to evaluate the two algorithms with synthetic data set generated from real world POIs and road networks.