Mining User Similarity from GPS Trajectory Based on Spatial-temporal and Semantic Information

In this study, we proposed a new framework to mine and analyze information from GPS trajectory data to find similar users from a spatial-temporal and semantic perspective. The framework combines spatial-temporal and semantic similarity techniques to achieve a system with low computational overhead and good similarity accuracy by using the characteristics of individual movements to identify similar users. It consists of three steps: first, spatial-temporal features are obtained by modeling and clustering stay points, and using them to calculate spatial-temporal similarities; next, using categories of points of interest within stay regions as semantic information, the semantic similarity can then be computed by frequent sequential pattern mining; finally, the spatial-temporal and semantic similarities can be combined to calculate the user similarity. We compared the results with those of related studies. The K-nearest neighbors experiments showed that the combination of spatial-temporal and semantic similarity methods exhibited excellent performance, being able to identify similar users more accurately. Consequently, our proposed method could be a useful identification framework in situations where large volumes of human spatial-temporal trajectory data exist, possibly due to the development of GPS devices and storage technology.