This is the source code from the ACL paper:

Jeremy Barnes, Roman Klinger, and Sabine Schulde im Walde. 2018. Bilingual Sentiment Embeddings: Joint Projection of Sentiment Across Languages. In Proceedings of ACL 2018.

If you use the code for academic research, please cite the paper in question:

@inproceedings{Barnes2018blse,
  author = 	"Barnes, Jeremy
		and Klinger, Roman
		and Schulte im Walde, Sabine",
  title = 	"Bilingual Sentiment Embeddings: Joint Projection of Sentiment Across Languages",
  booktitle = 	"Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
  year = 	"2018",
  publisher = 	"Association for Computational Linguistics",
  pages = 	"2483--2493",
  location = 	"Melbourne, Australia",
  url = 	"http://aclweb.org/anthology/P18-1231"
}

• Python 3
• NumPy
• sklearn [http://scikit-learn.org/stable/]
• pytorch [http://pytorch.org/]

First, clone the repo:

git clone https://github.com/jbarnesspain/blse
cd blse

Then, get monolingual embeddings, either by training your own, or by downloading the pretrained embeddings mentioned in the paper, unzipping them and putting them in the 'embeddings' directory:

Run the blse parameter search script, in order to get the best hyperparameters:

python3 parmeter_search.py

Finally, you can use the blse.py script which will automatically use the best hyperparameters found.

python3 blse.py

I have added further experiments comparing our method with MUSE. MUSE outperforms VecMap (version from 2017) on all setups and Barista on the binary setup. BLSE performs better than MUSE on 5 of the 6 setups.

Copyright (C) 2018, Jeremy Barnes

Licensed under the terms of the Creative Commons CC-BY public license