Clustering XML Documents Based on Data Type

The existing so-called semantic XML document clustering algorithms usually use a synonymous word library to calculate semantic similarities among XML documents. However, when people create their own XML documents, they name the element randomly and often use lots of abbreviations. Many tags are not real words at all. The XML documents created by different people may appear very different from each other even if they describe the same object. The traditional methods do not work well in such case. To address the problem, we proposed a novel similarity measure standard based on data-type tree, a model integrating data types and tags of XML documents. A clustering algorithm DT²K-means is also proposed to cluster XML documents. Empirical experiment results on real world data sets show DT²K-means can group the semantic similar XML documents together correctly, which contain different tags but describe the same object.