Acquisition of Named-Entity-Related Relations for Searching

Acquisition of Named-Entity-Related Relations for Searching* Tri-Thanh Nguyen and Akira Shimazu Japan Advanced Institute of Science and Technology 1-1, Asahidai, Nomi, Ishikawa, 923-1292, Japan {t-thanh,shimazu}@jaist.ac.jp Abstract. Named entities (NEs) are important in many Natural Language Processing (NLP) applications, and discovering NE-related relations in texts may be beneficial for these applications. This paper proposes a method to extract the ISA relation between a “named entity” and its category, and an IS-RELATED-TO relation between the category and its related object. Based on the pattern extraction algorithm “Person Category Extraction” (PCE), we extend it for solving our problem. Our experiments on Wall Street Journal (WSJ) corpus show promising results. We also demonstrate a possible application of these relations by utilizing them for semantic search. Keywords: Named-entity-related relations extraction, information extraction, pattern extraction, algorithm, semantic search. 1. Introduction Text documents often contain valuable relations among entities. For example, in the sentence taken from the WSJ corpus: There's a generally more positive attitude toward the economy, said Bette Raptapoulos, analyst for Prudential-Bache Securities Inc., … (1) there are relations: “Bette Raptapoulos” is-a analyst, and analyst for “Prudential-Bache Securities Inc.” Such relations may be beneficial in many NLP applications, such as for answering Who and List questions, e.g., “Who is Bette Raptapoulos?'' or “Give me the list of analyst for Prudential- Bache Securities Inc.” Relations in text documents can be extracted by pattern extraction as in (Brin 1998). (Brin 1998) presented Dual Iterative Pattern Relation Extraction (DIPRE), and used DIPRE to extract 〈author, title〉 tuples describing the relation: the author of the book title is author. Based on Brin's model, (Agichtein and Gravano 2000) presented the Snowball system to extract 〈organization, location〉 tuples indicating that the headquarters of organization is in location. (Nguyen and Shimazu 2007) developed the PCE system for extracting 〈person, category〉 tuples describing that the person is-a category. This study proposes to automatically extract quadruples 〈ne, category, related-to, object〉 describing that the named entity ne ISA category, and the category IS-RELATED-TO object. We call such relations “named-entity is-a category” relation, and “category related-to object” * This study was supported by Japan Advanced Institute of Science and Technology, the 21st Century COE Program: “Verifiable and Evolvable e-Society”. Copyright 2007 by Tri-Thanh Nguyen and Akira Shimazu. 349 relation. The related-to and object of a quadruple may be null. We extend PCE algorithm to extract quadruples, and build a semantic search system to utilize extracted quadruples for answering some types of questions. The remainder of this paper is organized as follows: Section 2 summarises some related work. Section 3 describes the original PCE algorithm and our extraction model. Section 4 gives a possible application of the extracted quadruples; Section 5 presents experiments and evaluation; Conclusions are given in the last section. 2. Related work (Brin 1998) presented the DIPRE algorithm for extracting relations, and used DIPRE to extract (author, title) tuples having the relation: the author of the book title is author. Starting with a small number of (author, title) seed tuples, DIPRE finds the occurrences …

Content truncated to 500 words