Information-Theoretic Approaches for Measuring the Structural Similarity of Semistructured Documents

We propose and experimentally evaluate different approaches for measuring the structural similarity of semistructured documents based on informationtheoretic concepts. Common to all approaches is a twostep procedure: first we extract and linearize the structural information from documents and then we use similarity measures that are based on, respectively, Kolmogorov complexity and Shannon entropy to determine the distance between the documents. Compared to other approaches, we are able to achieve a linear run time complexity and demonstrate in an experimental evaluation that the results of our technique in terms of clustering quality are on a par with or even better than those of other, slower approaches.