A Bayesian Generative Approach to Topic-Based Clustering in XML Corpora

This study explores XML partitioning through unsupervised topic modeling. We propose a novel mixed-membership Bayesian generative model for identifying latent topics within XML corpora. We derive approximate posterior inference and parameter estimation for the proposed XML topic model, implemented using a Gibbs sampling algorithm. This approach allows us to infer the topic distributions of the input XML documents, which are subsequently utilized to partition the entire XML corpus based on latent-topic similarity. Experiments conducted on real-world XML corpora demonstrate superior effectiveness compared to several methods.