Relation Extraction Using Support Vector Machine

R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 366 – 377, 2005. © Springer-Verlag Berlin Heidelberg 2005 Relation Extraction Using Support Vector Machine Gumwon Hong University of Michigan, Ann Arbor, MI 48109 gwhong@umich.edu Abstract. This paper presents a supervised approach for relation extraction. We apply Support Vector Machines to detect and classify the relations in Automatic Content Extraction (ACE) corpus. We use a set of features including lexical to- kens, syntactic structures, and semantic entity types for relation detection and classification problem. Besides these linguistic features, we successfully utilize the distance between two entities to improve the performance. In relation detec- tion, we filter out the negative relation candidates using entity distance thresh- old. In relation classification, we use the entity distance as a feature for Support Vector Classifier. The system is evaluated in terms of recall, precision, and F- measure, and errors of the system are analyzed with proposed solution. 1 Introduction The goal of Information Extraction (IE) is to pull out pieces of information that are salient to the user’s need from large volume of text. With the dramatic increase of World Wide Web, there has been growing interest in extracting relevant information from large textual documents. The IE tasks may vary in detail and reliability, but two subtasks are very common and closely related: named entity recognition and relation extraction. Named entity recognition identifies named objects of interest such as per- son, organizations or locations. Relation extraction involves the identification of ap- propriate relations among these entities. Examples of the specific relations are em- ployee-of and parent-of. Employee-of relation holds between a particular person and a certain organization and parent-of holds between a father and his child. According to Message Understanding Conferences (MUC), while named entities can be extracted with 90% or more in F measure with a state of the art system, relation extraction was not so successful [8]. In this paper we propose a supervised machine learning approach for relation ex- traction. The goal of this paper is to investigate an empirical method to find out useful features and experimental procedure to increase the performance of relation extrac- tion. We divide the extraction task into two individual subtasks: relation detection and relation classification. Relation detection is involved in identifying from every pair of entities positive examples of relations which can fall into one of many relation catego- ries. In relation classification, we assign a specific class to each detected relation. To each task, we apply distinct linguistic features ranging from lexical tokens to syntactic structures as well as the semantic type information of the entities. We also applied the distance between two entities to make the detection problem easier and to increase the Relation Extraction Using Support Vector Machine 367 performance of both the relation detection and classification. We apply Support Vec- tor Machines (SVMs) partly because it represents the state of the art performance for many classification tasks [3]. As we will discuss in section 3 about features, we are working on …

Content truncated to 500 words