Collective spatial keyword search on activity trajectories

Collective spatial keyword query (CSKQ) is one of the most useful spatial queries in location-based service systems. Although the availability of large-scale activity trajectories has given us useful knowledge of users' behavior, existing activity trajectory search methods are unable to support CSKQ queries reasonably. This paper studies effective and efficient CSKQ processing on activity trajectories to cover the gap. Specifically, we first formalize the problem by a trajectory based model that considers the spatial, activity and popularity issues, enabling more rational CSKQ results to be returned. To avoid high I/O cost, a novel hybrid index structure is further proposed to seamlessly integrate multi-domain information, so that inferior trajectories can be pruned during query processing. A novel candidate sub-trajectory search algorithm is also presented to reduce computation overhead by a linear scan on the trajectory. The experimental results on real check-in datasets demonstrate the efficiency and scalability of our proposed solution.