Mining Long, Sharable Patterns in Trajectories of Moving Objects

The efficient analysis of spatio–temporal data, generated by moving objects, is an essential requirement for intelligent location–based services. Spatiotemporal rules can be found by constructing spatio–temporal baskets, from which traditional association rule mining methods can discover spatio–temporal rules. When the items in the baskets are spatio–temporal identifiers and are derived from trajectories of moving objects, the discovered rules represent frequently travelled routes. For some applications, e.g., an intelligent ridesharing application, these frequent routes are only interesting if they are long and sharable, i.e., can potentially be shared by several users. This paper presents a database projection based method for efficiently extracting such long, sharable frequent routes. The method prunes the search space by making use of the minimum length and sharable requirements and avoids the generation of the exponential number of sub–routes of long routes. Considering alternative modelling options for trajectories, leads to the development of two effective variants of the method. SQL–based implementations are described, and extensive experiments on both real life– and large–scale synthetic data show the effectiveness of the method and its variants.