A Feature Selection Algorithm Based on Poisson Estimates

Feature selection is one of the key technologies for text categorization. Currently, it mainly includes technologies based statistics which is primarily from information theory and technologies based semantics which covers natural language processing, semantic Web etc.. Based on Poisson hypothesis, this article presents a new method combining both and tries to find features in documents with more semantic information. The contrast experiments carried on the Reuters-21578 corpus with the IG, Chi2 and WN algorithms show that this method has more advantages than other algorithms.