A clustering approach for XML linked documents

Clustering algorithms for hypertext documents consider not only the document content but also the links existing between them. All the similarity functions proposed in the literature assume that just one type of link exists between documents, with a unique semantic meaning. With the rapid diffusion of XML documents, a specific language, called XLink, has been proposed to specify inside XML documents different types of links. Each type of link forces a different degree of similarity between the documents on which it is defined, thus we claim it must influence in a different way the computation of distance values. In this paper, after presenting a graph-based formalization of the hypertexts we consider, we introduce a distance function, based on both the number and the type of the links connecting documents. Some preliminary experimental results on clustering algorithms based on the proposed function conclude the paper.