Automatic extraction of is-a relations in taxonomy learning

Abstract

Taxonomy learning is a prerequisite step for ontology learning. In order to create a taxonomy, first of all, existing 'is-a' relations between words should be extracted. A known way to extract 'is-a' relations is finding lexicosyntactic patterns in large text corpus. Although this approach produces results with high precision but it suffers from low values of recall. Furthermore developing a comprehensive set of patterns is tedious and cumbersome. In this paper, firstly, we introduce an approach for developing lexico-syntactic patterns automatically using the snippets of search engine results and then, challenge the law recall of this approach using a combined model, which is based on cooccurrence of pair words in the web and neural network classifier. Using our approach both precision and recall of extracted 'is-a' relations improved and FMeasure value reaches 0.72. © 2008 Springer-Verlag.