Discovering Relations among Named Entities from Large Corpora

Discovering Relations among Named Entities from Large Corpora Takaaki Hasegawa � Cyberspace Laboratories Nippon Telegraph and Telephone Corporation 1-1 Hikarinooka, Yokosuka, Kanagawa 239-0847, Japan hasegawa.takaaki@lab.ntt.co.jp Satoshi Sekine and Ralph Grishman Dept. of Computer Science New York University 715 Broadway, 7th floor, New York, NY 10003, U.S.A. � sekine,grishman � @cs.nyu.edu Abstract Discovering the significant relations embedded in documents would be very useful not only for infor- mation retrieval but also for question answering and summarization. Prior methods for relation discov- ery, however, needed large annotated corpora which cost a great deal of time and effort. We propose an unsupervised method for relation discovery from large corpora. The key idea is clustering pairs of named entities according to the similarity of con- text words intervening between the named entities. Our experiments using one year of newspapers re- veals not only that the relations among named enti- ties could be detected with high recall and precision, but also that appropriate labels could be automati- cally provided for the relations. 1 Introduction Although Internet search engines enable us to ac- cess a great deal of information, they cannot eas- ily give us answers to complicated queries, such as “a list of recent mergers and acquisitions of com- panies” or “current leaders of nations from all over the world”. In order to find answers to these types of queries, we have to analyze relevant documents to collect the necessary information. If many rela- tions such as “Company A merged with Company B” embedded in those documents could be gathered and structured automatically, it would be very useful not only for information retrieval but also for ques- tion answering and summarization. Information Ex- traction provides methods for extracting informa- tion such as particular events and relations between entities from text. However, it is domain depen- dent and it could not give answers to those types of queries from Web documents which include widely various domains. Our goal is automatically discovering useful re- lations among arbitrary entities embedded in large � This work is supported by Nippon Telegraph and Telephone (NTT) Corporation’s one-year visiting program at New York University. text corpora. We defined a relation broadly as an af- filiation, role, location, part-whole, social relation- ship and so on between a pair of entities. For ex- ample, if the sentence, “George Bush was inaugu- rated as the president of the United States.” exists in documents, the relation, “George Bush”(PERSON) is the “President of” the “United States” (GPE1), should be extracted. In this paper, we propose an unsupervised method of discovering relations among various entities from large text corpora. Our method does not need the richly annotated corpora required for supervised learning — corpora which take great time and effort to prepare. It also does not need any instances of relations as initial seeds for weakly supervised learning. This is an advan- tage of our approach, since we cannot know in ad- vance all the relations embedded in text. Instead, we only need a named entity (NE) …

Content truncated to 500 words