オントロジー主導による情報抽出 (「オントロジーの基礎と応用」)

概要

論文の詳細を見る
This paper proposes ontology-driven information extraction (ODIE), a novel approach to extracting information from a text corpus without using domain-dependent information extraction (IE) rules. Instead, the ODIE approach employs ontology as a semantic guide for selecting key information form the texts. First, we discuss a relation between information extraction and ontology, then we define application ontology for the purpose of ODIE. Next, we describe a novel method for achieving ODIE that consists of a marker passing method, the biasing of ralation paths, and the extraction of template slot names and fillers. Experimental results of the extraction of four pieces of information, such as company names and product names from 250 newspaper articles show high precision and recall. These results strangly support the feasibility and high potential of ODIE. We also discuss a possibility of a semi-automatic construction of the application ontology and the effect of incompleteness in the ontology on the IE task.
社団法人人工知能学会の論文
1999-11-01