Thursday, November 20, 2003

Jon Udell: Working with Bayesian Categorizers. [Scripting News] Multiclass problems are harder (more possible labels) than binary problems like spam filtering. In addition, there are many tricks for tweaking naïve Bayes (NB) classifiers; and other learning methods (SVM,boosting, winnow) may do significantly better than NB in document classification. It's nice to see new kinds of users exploring text-categorization technology. 8:41:13 PM