Domain concept handling in automated text categorization

Abstract

Single term based document representations, e.g. bag-of-words, have been widely accepted in the machine learning, information retrieval and text mining community. One notable limitation of such methods is that they do not consider the rich information resident in the semantic relations among terms. This paper reports our approach of concepts handling in document representation and its effect on the performance of text categorization. We introduce a Frequent word Sequence algorithm that generates concept-centered phrases to render domain knowledge concepts. Our experimental study based on a domain centered corpus shows that a consistent performance improvement can be achieved when concept-centered phrases are included in addition to the single term based features in document representations. We also observed that a universally suitable support threshold does not exist and the removal of concept irrelevant sequences can possibly further improve the performance at a lower support level.