Identifying main topics in density-based spatial clusters using network-based representative document extraction

Abstract

Geo-tagged documents on social media are usually related to local topics and events. Extracting areas of interest associated with local “attractive” topics from geo-tagged documents is one of the most important challenges in many application domains. In this paper, we propose a novel method for extracting the areas of interest from geo-tagged documents. There are two main steps in the proposed method. First, the (ε, σ)-density-based adaptive spatial clustering algorithm extracts areas where local topics are attracting attention as spatial clusters. Second, representative geo-tagged documents are detected to identify the main topic in each spatial cluster. The (ε, σ)-density-based adaptive spatial clustering algorithm changes the threshold for seamlessly extracting spatial clusters regardless of the local densities of the posted geo-tagged documents. Moreover, the proposed method utilizes the network-based important sentence extraction method in order to extract representative geo-tagged documents from each spatial cluster. The experimental results show that the proposed method can extract the areas of interest as spatial clusters and representative documents as main topics.