DOCUMENT CLUSTERING USING PROBABILISTIC NETWORKS

Abstract

We use Bayesian Belief Networks to perform inference on the similarity between a document and a cluster. We introduce the idea of "similar region" to construct the Bayesian Belief Netwoks and use a heuristic method to select initial seed objects. Our method allows a document to either belong to exactly one cluster or more than one cluster. Experiments are conducted to compare our method against K-means clustering method on a set of one-line Chinese news extracted from internet.