A Framework for Comparing Text Categorization Approaches

Machine Learning in Information Access

Volume

Papers from the 1996 AAAI Spring Symposium

Contents

For the past few years, text categorization has emerged as an application domain to machine learning techniques. Several approaches have already been proposed. This paper does not present yet another technique. It is rather an attempt to unify the approaches encountered so far. Moreover this state-ofthe- art enables us to stress a shortcoming in earlier research: the lack of evaluation of inductive learners in the categorization process. We present a first attempt to remedy this lack. We expose an experimental framework, that fits in with our unified view of text categorization methods. This framework allows us to conduct a set of tentative experiments in order to assess which characteristics allow a learner to perform well on the text categorization task.