GPT-Negochat: A more realistic corpus of negotiation dialogues
The GPT-Negochat corpus is a modified version of the original Negochat corpus, which contains negotiation dialogues between an Employer and a Candidate. The utterances in the original corpus were generated using a template-based NLG module and therefore, sound robotic and in general, do not sound convincingly real.
GPT-Negochat is the result of using GPT-3 to modify this original corpus to make the dialogues resemble actual job-negotiation dialogues more closely while still retaining the original meaning of the utterances.
In addition to rephrasing the utterances, a small set of highly unrealistic dialogue segments have been removed in GPT-Negochat without affecting the coherence of the surrounding dialogue.
Reference(s)
If you're using this dataset, please cite this repository and its owner, Amogh Mannekote.
Please also cite the following original work:
@inproceedings{konovalov2016negochat,
title={The Negochat Corpus of Human-agent Negotiation Dialogues},
author={Konovalov, Vasily and Artstein, Ron and Melamud, Oren and Dagan, Ido},
booktitle={Proceedings of the 10th International Conference on Language Resources and Evaluation, LREC 2016},
year={2016},
pages={3141-3145},
url={www.lrec-conf.org/proceedings/lrec2016/pdf/240_Paper.pdf}
}