Learning Dictionaries for Named Entity Recognition using Minimal Supervision

Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 452–461, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics Learning Dictionaries for Named Entity Recognition using Minimal Supervision Arvind Neelakantan Department of Computer Science University of Massachusetts, Amherst Amherst, MA, 01003 arvind@cs.umass.edu Michael Collins Department of Computer Science Columbia University New-York, NY 10027, USA mcollins@cs.columbia.edu Abstract This paper describes an approach for au- tomatic construction of dictionaries for Named Entity Recognition (NER) using large amounts of unlabeled data and a few seed examples. We use Canonical Cor- relation Analysis (CCA) to obtain lower dimensional embeddings (representations) for candidate phrases and classify these phrases using a small number of labeled examples. Our method achieves 16.5% and 11.3% F-1 score improvement over co-training on disease and virus NER re- spectively. We also show that by adding candidate phrase embeddings as features in a sequence tagger gives better perfor- mance compared to using word embed- dings. 1 Introduction Several works (e.g., Ratinov and Roth, 2009; Co- hen and Sarawagi, 2004) have shown that inject- ing dictionary matches as features in a sequence tagger results in significant gains in NER perfor- mance. However, building these dictionaries re- quires a huge amount of human effort and it is of- ten difficult to get good coverage for many named entity types. The problem is more severe when we consider named entity types such as gene, virus and disease, because of the large (and growing) number of names in use, the fact that the names are heavily abbreviated and multiple names are used to refer to the same entity (Leaman et al., 2010; Dogan and Lu, 2012). Also, these dictionaries can only be built by domain experts, making the pro- cess very expensive. This paper describes an approach for automatic construction of dictionaries for NER using large amounts of unlabeled data and a small number of seed examples. Our approach consists of two steps. First, we collect a high recall, low preci- sion list of candidate phrases from the large unla- beled data collection for every named entity type using simple rules. In the second step, we con- struct an accurate dictionary of named entities by removing the noisy candidates from the list ob- tained in the first step. This is done by learning a classifier using the lower dimensional, real-valued CCA (Hotelling, 1935) embeddings of the can- didate phrases as features and training it using a small number of labeled examples. The classifier we use is a binary SVM which predicts whether a candidate phrase is a named entity or not. We compare our method to a widely used semi- supervised algorithm based on co-training (Blum and Mitchell, 1998). The dictionaries are first evaluated on virus (GENIA, 2003) and disease (Dogan and Lu, 2012) NER by using them directly in dictionary based taggers. We also give results comparing the dictionaries produced by the two semi-supervised approaches with dictionaries that are compiled manually. The effectiveness of the dictionaries are …

Content truncated to 500 words