Exploiting Domain Structure for Named Entity Recognition

Proceedings of the Human Language Technology Conference of the North American Chapter of the ACL, pages 74–81, New York, June 2006. c⃝2006 Association for Computational Linguistics Exploiting Domain Structure for Named Entity Recognition Jing Jiang and ChengXiang Zhai Department of Computer Science University of Illinois at Urbana-Champaign Urbana, IL 61801 {jiang4,czhai}@cs.uiuc.edu Abstract Named Entity Recognition (NER) is a fundamental task in text mining and nat- ural language understanding. Current ap- proaches to NER (mostly based on super- vised learning) perform well on domains similar to the training domain, but they tend to adapt poorly to slightly different domains. We present several strategies for exploiting the domain structure in the training data to learn a more robust named entity recognizer that can perform well on a new domain. First, we propose a sim- ple yet effective way to automatically rank features based on their generalizabilities across domains. We then train a classifier with strong emphasis on the most general- izable features. This emphasis is imposed by putting a rank-based prior on a logis- tic regression model. We further propose a domain-aware cross validation strategy to help choose an appropriate parameter for the rank-based prior. We evaluated the proposed method with a task of recog- nizing named entities (genes) in biology text involving three species. The exper- iment results show that the new domain- aware approach outperforms a state-of- the-art baseline method in adapting to new domains, especially when there is a great difference between the new domain and the training domain. 1 Introduction Named Entity Recognition (NER) is the task of identifying and classifying phrases that denote cer- tain types of named entities (NEs), such as per- sons, organizations and locations in news articles, and genes, proteins and chemicals in biomedical lit- erature. NER is a fundamental task in many natural language processing applications, such as question answering, machine translation, text mining, and in- formation retrieval (Srihari and Li, 1999; Huang and Vogel, 2002). Existing approaches to NER are mostly based on supervised learning. They can often achieve high accuracy provided that a large annotated training set similar to the test data is available (Borthwick, 1999; Zhou and Su, 2002; Florian et al., 2003; Klein et al., 2003; Finkel et al., 2005). Unfortunately, when the test data has some difference from the training data, these approaches tend to not perform well. For ex- ample, Ciaramita and Altun (2005) reported a per- formance degradation of a named entity recognizer trained on CoNLL 2003 Reuters corpus, where the F1 measure dropped from 0.908 when tested on a similar Reuters set to 0.643 when tested on a Wall Street Journal set. The degradation can be expected to be worse if the training data and the test data are more different. The performance degradation indicates that exist- ing approaches adapt poorly to new domains. We believe one reason for this poor adaptability is that these approaches have not considered the fact that, depending on the genre or domain of the text, the entities to be …

Content truncated to 500 words