Information extraction by text classification

Information extraction and text classification are usually seen as complementary forms of shallow text processing, in that they are aimed at very different tasks. In this paper, we describe two simple but real-world domains in which text classification techniques can be used directly for information extraction. Specifically, we describe systems for extracting information from business cards, and for automatically processing “change of address” email messages, that are based primarily on text classification techniques. Our main technical contribution is a novel integration of hidden Markov models and text classifiers.