Proceedings of the Pacific Knowledge Acquisition Workshop 2004

This paper investigates whether and how natural language processing and data mining techniques can be utilized for locating desired knowledge in a large text collection. This task amounts to finding cue words and phrases indicating the location of knowledge, where the challenge is to establish a methodology that can cope with the diversity of expressions. We examine the feasibility of mining cue expressions from the syntactic dependency structure obtained from parsed sentences. As a case study, the (phrasal) expressions concerning a variety of tests related to chronic hepatitis were sought in the Medline abstracts. We observed that dependency analysis helped to narrow down the candidates for verbal expressions, although it was ineffective for other types of expressions.