Mining Frequent Semantic Regions from Low-Sampling-Rate Data

Abstract

Owing to the pervasive of GPS-equipped mobile phones today, the locations of user can be easily determined and collected. As location data provide very valuable information that can be useful for various location-based services, semantic region mining becomes a hot issue in recent years. In this paper, we propose a new framework SeReMine for mining less energy-consuming data, also termed as low-sampling-rate data, to discover semantic regions of a user. We assert that a daily motion would leave some clues with another one if they follow the same movement behavior, and semantic regions can be inferred from these movement behavior. To extract clues carefully, apart from distance, various data types, occurrence time and user preference also need to be taken into consideration. Based on this assertion, we propose clue-aware referency to measure the clues among daily motions. Accordingly, we utilize clue-aware clustering [6] to cluster daily motions which have mutually high referency into groups to capture the movement behaviors. Finally, we devise the region screening to examine each cluster in order to determine the final semantic regions. We validate our ideas and evaluate the proposed framework by experimenting on synthetic datasets and the results demonstrate that our SeReMine are more effective than the other techniques for mining semantic regions.

Owing to the pervasive of GPS-equipped mobile phones today, the locations of user can be easily determined and collected. As location data provide very valuable information that can be useful for various location-based services, semantic region mining becomes a hot issue in recent years. In this paper, we propose a new framework SeReMine for mining less energy-consuming data, also termed as low-sampling-rate data, to discover semantic regions of a user. We assert that a daily motion would leave some clues with another one if they follow the same movement behavior, and semantic regions can be inferred from these movement behavior. To extract clues carefully, apart from distance, various data types, occurrence time and user preference also need to be taken into consideration. Based on this assertion, we propose clue-aware referency to measure the clues among daily motions. Accordingly, we utilize clue-aware clustering [6] to cluster daily motions which have mutually high referency into groups to capture the movement behaviors. Finally, we devise the region screening to examine each cluster in order to determine the final semantic regions. We validate our ideas and evaluate the proposed framework by experimenting on synthetic datasets and the results demonstrate that our SeReMine are more effective than the other techniques for mining semantic regions.