Hotspots Detection on Microblog

Abstract

Main microblog research is focus on the structural analysis of social networks, rather than the text and topic analysis. Traditional topic detection methods could not be applied due to the microblog short text features and structural characteristics. We taken advantage of availability of latent Dirichlet allocation (LDA) to expand the text feature space, and used frequency statistics for our topic ranking, and improved it based on the microblog nontext element data and word element. We taken into account both the text context similarity and semantic similarity in order to make it possible that the traditional clustering method can make difference in the microblog text topic analysis. Experimental studies show our method works well on microblog dataset.