Hybrid clustering for identification of distinct topics of a domain using user influence pattern

Abstract

Content based tweet clustering is extensively used for automatic topic identification of tweets in social media analytics. However due to restrictions on the length of the content in social media platforms like twitter mere content is not enough to provide sufficient information for clustering. In this paper the authors proposed to enhance the clustering quality by adding tweeting behavior of influential users. Spearmen correlation is appropriately adapted for identifying mergeable clusters. A new methodology for hybrid clustering is proposed and tested using entropy on real data related to three domains namely sports, politics, and health. The proposed method achieved distinct cluster formation which is reflected by reduced entropy after applying merging based on user influence patterns.