XML Document Clustering

The ability of providing a “standardized, extensible means of coupling semantic information within documents describing semistructured data” (Chaudhri, Rashid, & Zicari, 2003) has led to a steady growth of XML (extensible markup language) data sources, so that XML is touted as the driving force for representing and exchanging data on the Web. The motivation behind any clustering problem is to find an inherent structure of relationships in the data and expose this structure as a set of clusters where the objects within the same cluster are each to other highly similar but very dissimilar from objects in different clusters. The clustering problem finds in text databases a fruitful research area. Since today semistructured text data has become more prevalent on the Web, and XML is the de facto standard for such data, clustering XML documents has increasingly attracted great attention. Any application domain that needs organization of complex document structures (e.g., hierarchical structures with unbounded nesting, object-oriented hierarchies) as well as data containing a few structured fields together with some largely unstructured text components can be profitably assisted by an XML document clustering task. In principle, the availability of schemas for XML data may be useful to drive or simplify a clustering task. For instance, in case of structural classification, XML documents with different element values but similar schemas could be grouped together. An XML DTD defines the document schema by means of constraints (element content models) that specify the element types, hierarchical relationships between elements, and other properties such as multiple occurrences of elements (operator +), optional elements (operator ?), and alternate elements (operator |). XML documents available from real data sources tend to have such characteristics. However, exploiting XML schemas profitably for classification purposes is not always feasible in practice. On one hand, most XML sources provide documents that are schema-less, that is, documents without an explicitly associated element type definition. On the other hand, XML documents available from the same data source may have quite different size and structure mainly due to nesting and repetition of elements, although they conform to a unique DTD. Also, XML documents with different schemas may have similar contents, or, in a more complicated case, XML documents coming from heterogeneous sources may represent semantically related data even if Chapter LXXI XML Document Clustering