The decision list algorithm is one of the most successful algorithms for classification problems in natural language processing. We propose a method based on Bayesian learning to calculate the reliability of contextual evidences in decision lists. The method also gives well-founded smoothing and better use of prior information of each type of contextual evidence. We evaluate these proposed methods on Japanese word sense disambiguation problems. The results show improved accuracy close to one expected from the Bayesian theory.
展开▼
