Discovering Personal Paths from Sparse GPS Traces

Personal paths capture “personal meaningful places” [13, 14] in temporal sequence. Knowledge of a user’s paths enables novel and useful features for location-aware applications, e.g., traffic condition updates for commuting routes, carpool partner finding with similar commuting routes and schedules. Prior work has explored algorithms to discover “significant locations” [1,2] and “transportation routes” [6, 9] from GPS data, however, we know of no algorithms specifically designed for sparse GPS traces, which represent typical location datasets collected from GPS enabled mobile devices. In this paper, we report two spatio-temporal clustering algorithms, TDJ and R-TDJ, for discovering personal paths. Specifically, the algorithms are designed to address the noisy and sparse nature of GPS data. Our experiment results show that both TDJ and R-TDJ discovered meaningful spatio-temporal clusters to form personal paths. R-TDJ demonstrates better performance on sparse GPS data.