XML document clustering using structure-preserving flat representation of XML content and structure

Abstract

With the increasing use of XML in many domains, XML document clustering has been a central research topic in semistructured data management and mining. Due to the semistructured nature of XML data, the clustering problem becomes particularly challenging, mainly because structural similarity measures specifically designed to deal with tree/graph-shaped data can be quite expensive. Specialized clustering techniques are being developed to account for this difficulty, however most of them still assume that XML documents are represented using a semistructured data model. In this paper we take a simpler approach whereby XML structural aspects are extracted from the documents to generate a flat data format to which well-established clustering methods can be directly applied. Hence, the expensive process of tree/graph data mining is avoided, while the structural properties are still preserved. Our experimental evaluation using a number of real world datasets and comparing with existing structural clustering methods, has demonstrated the significance of our approach.