Current work on the project is focusing on developing an improved Bayesian classification model and developing new approaches to active learning with a Bayesian model. ? 1) We have found through extensive testing that our version of naive Bayes, a form of MBM (multivariate Bernoulli model), is at least as effective as the MM (multinomial model). The MM model attempts to extract information from local feature counts in text documents. We have developed what we call a Stacked MBM model, which shows that there is not sufficient independent information in the local counts to make a significant improvement in performance. ? 2) We have developed term based active learning methods which provide a different approach to active learning and have shown that they are in many cases more effective then simple uncertainty sampling or error reduction sampling.? 3) We have developed an example selection method that is very powerful in improving Bayes on all of MEDLINE. This is important because there are few methods that can really be applied to all of MEDLINE.
