Using semantic relatedness measures with dynamic self-organizing maps for improved text clustering

Growing volumes of text and increasing expectations on the complexity of analysis entail advanced approaches to text mining. Unsupervised text clustering is an efficient approach to determine structural groupings in a text corpus without the impact of external bias. The information content of such structural groupings needs to be enhanced by integrating semantics into the cluster outcomes. This integration can eventuate at different stages of the clustering process where semantic relatedness measures can be used for the integration. In this paper we propose a novel method of semantics integration to cluster separation in a dynamic self-organizing map algorithm. We demonstrate the effectiveness of the proposed method and the value of semantics integration for cluster separation with empirical results from two benchmark text datasets.