Learning rules with negation for text categorization

This paper describes Olex, a novel method for the automatic construction of rule-based text classifiers. Olex relies on an optimization algorithm whereby a set of (both positive and negative) discriminating terms is generated for the category being learned. Such terms are then used to construct a classifier of the form "if term t"1 or ... term tn occurs in document d, and none of terms tn - 1, tn - m occurs in d, then d belongs to category c". The proposed method is simple and elegant. Despite this, the results of a systematic experimentation performed on both the REUTERS-21578 and the OHSUMED data collections show that Olex is both effective and efficient.