Spectral clustering and community detection in document networks

Abstract

Document clustering is one of the most active research topics in text mining. In this work two approaches issued from very different fields are explored for document clustering: spectral clustering and community detection in complex networks. Both approaches are based on a representation of the document collection as a graph, of which the nodes represent the documents and the edges represent the similarities between each pair of documents, such that the two approaches have many issues in common. The results of the application of these two types of techniques to benchmark text mining problems show that they are complementary and are useful for finding structure in large collections of documents