A Semantic Feature for Relation Recognition Using a Web-based Corpus

A Semantic Feature for Relation Recognition Using A Web-based Corpus Chen-Ming Hung Institute of Information Science Academia Sinica, Taipei, Taiwan rglly@iis.sinica.edu.tw Abstract Selecting appropriate features to represent an entity pair plays a key role in the task of relation recognition. However, existing syntactic features or lexical features cannot capture the interaction between two enti- ties because of the dearth of annotated rela- tional corpus specialized for relation recog- nition. In this paper, we propose a seman- tic feature, called the latent topic feature, which is topic-based and represents an en- tity pair at the semantic level instead of the word level. Moreover, to address the prob- lem of insufficiently annotated corpora, we propose an algorithm for compiling a train- ing corpus from the Web. Experiment results demonstrate that latent topic features are as effective as syntactic or lexical features. Moreover, the Web-based corpus can resolve the problems caused by insufficiently anno- tated relational corpora. 1 INTRODUCTION Relation recognition is a challenging task because finding appropriate features to represent the rela- tionship between two entities is difficult and limited by the scarcity of annotated corpora. Prior works on relation recognition have focused on syntactic features, e.g., parsing trees (Culotta and Sorensen, 2004; Zelenko et al., 2003), and on lexical fea- tures, e.g., Part-Of-Speech (POS) features. These approaches show that syntactic features and lexical features outperform bag-of-words (BOW) on exist- ing annotated corpora such as the RDC corpus of the ACE project. The superior performance achieved by syntactic and lexical features is due to their ability to capture the grammatical relations between two entities and the characteristics of the entities. For example, (Culotta and Sorensen, 2004) add hyper- nyms of entities to features derived from WordNet. However, neither syntactic nor lexical features can capture the interaction between two entities at the semantic level. Another issue in the task of relation recognition is insufficiently annotated corpora. For example, given a pair {the U.N. body, Kosovo}, we can only find three sentences containing both entities in the RDC corpus, which is commonly used corpus in the relation recognition task. The problem of an in- sufficiently annotated corpus biases feature vectors and distorts the prediction of entity pairs. How- ever, (Huang et al., 2004; Hung and Chien, 2007) have shown that the Web can be used as an alterna- tive source of documents related to a given query. That is possibly because of the increasing size of the Web and the efficiency in commercial search en- gines, e.g., Google and Yahoo!. To resolve the above problems, we propose a se- mantic feature called the latent topic feature, which is extracted by exploiting the Latent Dirichlet Al- location (LDA) algorithm. Unlike syntactic fea- tures or lexical features, latent topic features repre- sent entity pairs as random mixtures of latent topics, where each topic is characterized by a distribution of words. We prove experimentally that latent topic features are as effective as syntactic features or lexi- cal features in capturing the interaction between two entities. The experiment …

Content truncated to 500 words