EBSCAN: An Entanglement-based Algorithm for Discovering Dense Regions in Large Geo-social Data Streams with Noise

Abstract

The remarkable growth of social networking services on global positioning system (GPS)-enabled handheld devices has produced enormous amounts of georeferenced big data. Given a large spatial dataset, the challenge is to effectively discover dense regions from the dataset. Dense regions might be the most attractive area in a city or the most dangerous zone of a town. A solution to this problem can be useful in many applications, including marketing, tourism, and social research. Density-based clustering methods, such as DBSCAN, are often used for this purpose. Nevertheless, current spatial clustering methods emphasize density while neglecting human behavior derived from geographical features. In this paper, we propose EBSCAN, which is based on the novel idea of an entanglement-based approach. Our method considers not only spatial information but also human behavior derived from geographical features. Another problem is that competing methods such as DBSCAN have two input parameters. Thus, it is difficult to determine optimal values. EBSCAN requires only a single intuitive parameter, tooFar, to discover dense regions. Finally, we evaluate the effectiveness of the proposed method using both toy examples and real datasets. Our experimentally obtained results reveal the properties of EBSCAN and show that it is >10 times faster than the competitor.