Discovering Relations among Named Entities by Detecting Community Structure

Discovering Relations among Named Entities by Detecting Community Structure Tingting He1 2 ,Junzhe Zhao1 ,Jing Li1 1 Department of Computer Science, Huazhong Normal University 430079 Wuhan, China 2 Software College of Tsinghua University 102201 Beijing, China tthe@mail.ccnu.edu.cn zhaojunzhe@mails.ccnu.edu.cn lee_king8@hotmail.com Abstract. This paper proposes a networked data mining method for relations discovery from large corpus. The key idea is representing the named entities pairs and their contexts as the network structure and detecting the communities from the network. Then each community relates to a relation the named entities pairs in the same community have the same relation. Finally, we labeled the relations. Our experiment using the corpus of People's Daily reveals not only that the relations among named entities could be detected with high precision, but also that appropriate labels could be automatically provided for the relations. Keywords: named entities pair, community, betweenness 1 Introduction Relation extraction among named entities (NR) is one of major tasks in information extraction (IE).The goal of relation extraction is to find out the relations among named entities(NE) in documents. In recent years, such technology has been widely used in many fields, such as: information retrieval, question- answering systems, biology technology, construction of ontology, etc. Many methods have been proposed for relation extraction, including supervised learning methods (Zelenko et al., 2002) [8], weakly supervised learning methods (Brin 1998; Agichtein 2000; Sudo 2003) [9][10][11] and unsupervised learning methods (Hasegawa 2004; Chen Jinxiu 2005) [1][2]. In this paper, we propose an unsupervised method for relation extraction. Currently, unsupervised learning methods for relation extraction have some difficulties. For example, in Hasegawa et al.'s method, they eliminated less frequent NE pairs, collected the contexts of NE pairs, clustered the contexts using complete linkage method, and finally selected the most frequent word in a cluster to label the relation in this cluster. However, the less frequent NE pairs might have relations, and it is difficult for complete linkage method to select the threshold. In Chen Jinxiu et al.'s method, they firstly got the NE pairs labeled with relation in ACE corpus and collected their contexts, clustered the contexts using stability-based method, and finally used DCM method to label the clusters. However, we can't select NE pairs with relation in the unannotated corpus. And the relations of NE pairs might be in hierarchical structure; but the stability-based method can not discover the hierarchical structure. Based on the issues mentioned above, this paper proposes a method to discover relations among NEs based on networked data mining. The advantages of this method is that there is no need to eliminate less frequent NE pairs and select the NE pairs which have relations. In addition, our method can automatically present the relations of NE pairs in the hierarchical structure. The rest of this paper is organized as follows. Section 2 talks about the construction for networked structure. Section 3 explores the procedure of relation discovering and labeling. Section 4 describes experiments and evaluation of experiment results. Section 5 gives a discussion of the problems in …

Content truncated to 500 words