Title

A Domain Adaptive Approach to Automatic Acquisition of Domain Relevant Terms and their Relations with Bootstrapping

Authors

Feiyu Xu (DFKI – German Research Center for Artificial Intelligence Stuhlsatzenhausweg 3, 66 123 Saarbrücken, Germany)

Daniela Kurz (XtraMind GmbH Stuhlsatzenhausweg 3, 66 123 Saarbrücken, Germany)

Jakub Piskorski (DFKI – German Research Center for Artificial Intelligence Stuhlsatzenhausweg 3, 66 123 Saarbrücken, Germany)

Sven Schmeier (XtraMind GmbH Stuhlsatzenhausweg 3, 66 123 Saarbrücken, Germany)

Session

WO3: Acquisition Of Lexical Information

Abstract

In this paper, we present an unsupervised hybrid text-mining approach to automatic acquisition of domain relevant terms and their relations. We deploy the TFIDF-based term classification method to acquire domain relevant single -word terms. Further, we apply two strategies in order to learn lexico-syntatic patterns which indicate paradigmatic and domain relevant syntagmatic relations between the extracted terms. The first one uses an existing ontology as initial knowledge for learning lexico-syntactic patterns, while the second is based on different collocation acquisition methods to deal with the free-word order languages like German. This domain-adaptive method yields good results even when trained on relatively small training corpora. It can be applied to different real-world applications, which need domain-relevant ontology, for example , information extraction, information retrieval or text classification.

Keywords

Bootstrapping

Full Paper

351.pdf