##Embedding Features for Semi-supervised Learning Author: Jiang Guo, jguo@ir.hit.edu.cn

This tool is used for semi-supervised learning of NER, using various kinds of embedding features. This tool is associated with (Guo et al., 2014). The proposed approaches are shown to be much better than the direct usage of continuous word embedding features.

###Requirements:

• CRFSuite

###Data you need:

• CoNLL-2003 NER dataset

The original dataset should be converted to BIO-style annotation, see data/ner/get_data.md .

Before run, check the data/ner/bin/make.eng.2016 tool (code modification required) to generate necessary training and test data !!!

###Training an NER tagger

$ ./train.sh [de|bi|ce|proto]

• de - dense embedding features
• bi - binarized embedding features
• ce - clustered features
• proto - distributional prototype features

To use the combined features, e.g. de+proto

###Testing

$ ./tag.sh [de|bi|ce|proto]

###Reference

@InProceedings{guo-EtAl:2014:EMNLP2014,
  author    = {Guo, Jiang  and  Che, Wanxiang  and  Wang, Haifeng  and  Liu, Ting},
  title     = {Revisiting Embedding Features for Simple Semi-supervised Learning},
  booktitle = {Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  month     = {October},
  year      = {2014},
  address   = {Doha, Qatar},
  publisher = {Association for Computational Linguistics},
  pages     = {110--120},
  url       = {http://www.aclweb.org/anthology/D14-1012}
}