ljynlp/hitrans

HiTrans: A Transformer-Based Context- and Speaker-Sensitive Model for Emotion Detection in Conversations

Source code for COLING 2020 paper: HiTrans: A Transformer-Based Context- and Speaker-Sensitive Model for Emotion Detection in Conversations

1. Environments

• python (3.6.8)
• cuda (10.1)

2. Dependencies

• numpy (1.17.3)
• torch (1.6.0)
• transformers (3.5.1)
• pandas (1.1.5)
• scikit-learn (0.23.2)

3. Preparation

• Download MELD dataset
• Put train_sent_emo.csv, dev_sent_emo.csv and test_sent_emo.csv into the directory data/

4. Training

>> python main.py

5. Evaluation

>> python main.py --evaluate

6. License

This project is licensed under the MIT License - see the LICENSE file for details.

7. Citation

If you use this work or code, please kindly cite the following paper:

@inproceedings{li-etal-2020-hitrans,
    title = "{H}i{T}rans: A Transformer-Based Context- and Speaker-Sensitive Model for Emotion Detection in Conversations",
    author = "Li, Jingye  and
      Ji, Donghong  and
      Li, Fei  and
      Zhang, Meishan  and
      Liu, Yijiang",
    booktitle = "Proceedings of the 28th International Conference on Computational Linguistics",
    year = "2020",
    publisher = "International Committee on Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.coling-main.370",
    pages = "4190--4200",
}

hitrans's Issues

请问可以自己预训练吗

EmoryNLP dataset statistics

For EmoryNLP dataset, why the utterance count 7551/954/984 reported in the paper is different from 9934/1344/1328 as in EmoryNLP official repository?

PLZ! For the help

Hi, I am trying to reproduce your paper recently. However, I can not get the same performance in my trial. And could you free your code or some useful details for me? thx