Extrofitting: Enriching Word Representation and its Vector Space with Semantic Lexicons

Post-processing method for enriching word representation and its vector space.

Extrofitting outperforms Faruqui's retrofitting with generalization.

Paper Links: https://arxiv.org/abs/1804.07946

Extrofitted GloVe (glove.42B.300d) can be downloaded at this url

: https://drive.google.com/open?id=1R82TOyAmOgaAjCVrvUfBlJPJaTD5x28a

Requirements

• Jupyter notebook with Python 2.7
• numpy
• sklearn
• tqdm (optional)

Downloads

• pretrained simple GloVe is borrowed from counter-fitting (https://github.com/nmrksic/counter-fitting)

  • Download at https://drive.google.com/file/d/1x7Ch3uC8U2a1D9rab601tTPdyxoBg1wp/view?usp=sharing

• Codes are based on retrofitting (https://github.com/mfaruqui/retrofitting)

• Semantic lexicons are also borrowed from retrofitting (included in this GitHub)

• MEN-3k dataset is from (https://staff.fnwi.uva.nl/e.bruni/MEN)

Reference

If you use this algorithm on your pretrained word vector, please cite the following paper:

@article{jo2018extrofitting,
  title={Extrofitting: Enriching Word Representation and its Vector Space with Semantic Lexicons},
  author={Jo, Hwiyeol and Choi, Stanley Jungkyu},
  journal={arXiv preprint arXiv:1804.07946},
  year={2018}
}