Topic discovery from document using ant-based clustering combination

Abstract

This paper presents a topic discovery approach based on multi-ant colonies clustering combination. The algorithm consists of three parts. First, each document is represented as a vector of features in a vector space model. Then a hypergraph model is used to combine the clusterings produced by three kinds of ant-based algorithms with different moving speed. Finally, the topic of each cluster is extracted by re-computing the term weights. Test results show that the number of topics can be adaptively determined and clustering combination can improve the system performance. © Springer-Verlag Berlin Heidelberg 2005.