Combining structure and content similarities for XML document clustering

Abstract

This paper proposes a clustering approach that explores both the content and the structure of XML documents for determining similarity among them. Assuming that the content and the structure of XML documents play different roles and importance depending on the use and purpose of a dataset, the content and structure information of the documents are handled using two different similarity measuring methods. The similarity values produced from these two methods are then combined with weightings to measure the overall document similarity. The effect of structure similarity and content similarity on the clustering solution is thoroughly analysed. The experiments prove that clustering of the text-centric XML documents based on the content-only information produces a better solution in a homogeneous environment, documents that derived from one structural definition; however, in a heterogeneous environment, documents that derived from two or more structural definitions, clustering of the text-centric XML documents produces a better result when the structure and the content similarities of the documents are combined with different strengths.