Learning to Classify English Text with Ilp Methods

Text categorization is the task of classifying text into one of several pre-deened categories. In this paper we will evaluate the eeectiveness of several ILP methods for text categorization, and also compare them to their propositional analogs. The methods considered are FOIL, the propositional rule-learning system RIPPER, and a rst-order version of RIPPER called FLIPPER. We show that the beneet of using a rst-order representation in this domain is relatively modest; in particular, the performance diierence between FLIPPER and FOIL and their propositional counterparts is quite small, compared to the diierences between FOIL and FLIPPER. However, a rst-order representation seems to be advantageous when high-precision classiiers are desirable.