Mining frequent spatial-textual sequential patterns

Abstract

Penetration of GPS-enabled devices has resulted into generation of a lot of Spatial-Textual data, which can be mined/analyzed to improve various location-based services. One such kind of data is activity-trajectory data, i.e. a sequence of locations visited by a user with each location having a set of activities performed by the user. In this thesis, we propose a mining framework along with algorithms for mining activity-trajectory data to nd out Spatial-Textual sequencial patterns. The proposed framework is exible in the sense that any algorithm from the existing sequence mining algorithms can be used as a core algorithm in our framework. We design and implement three di erent algorithms, namely, Spatial-Textual sequence mining algorithm, Textual-Spatial sequence mining algorithm and Hybrid sequence mining algorithm and nd out their e ectiveness for di erent location granularity and sensitivities. The experiment results shows Spatial-Textual approach outperforming other approaches in case of better location selectivity in the data. We also observe that the Spatial-Textual approach is able to handle much larger activity-trajectory data as compared to other approaches.