Word Sense Disambiguation for Automatic Taxonomy Construction from Text-Based Web Corpora

摘要

In this paper, we propose the Automatic Taxonomy Construction from Text (ATCT) framework for building taxonomies from text-based Web corpora. The framework is composed of multiple processing steps. Firstly, domain terms are extracted using a filtering method. Subsequently, Word Sense Disambiguation (WSD) is optionally applied in order to determine the senses of these terms. Then, by means of a subsumption technique, the resulting concepts are arranged in a hierarchy. We construct taxonomies with and without WSD and we investigate the effect of WSD on the quality of concept type-of relations using an evaluation framework that uses a golden taxonomy. We find that WSD improves the quality of the built taxonomy in terms of the taxonomic F-Measure.