Learning hidden markov model topology based on KL divergence for information extraction

Abstract

To locate information embedded in documents, information extraction systems based on rule-based pattern matching have long been used. To further improve the extraction generalization, hidden Markov model (HMM) has recently been adopted for modeling temporal variations of the target patterns with promising results. In this paper, a state-merging method is adopted for learning the topology with the use of a localized Kullback Leibler (KL) divergence. The proposed system has been applied to a set of domain-specific job advertisements and preliminary experiments show promising results.