himanshu03vsk/eacl2021-OffensEval-Dravidian

EACL 2021 paper (SJ_AJ@DravidianLangTech-EACL2021: Task-Adaptive Pre-Training of Multilingual BERT models for Offensive Language Identification)

EACL 2021 OffensEval in Dravidian Languages

Leaderboard (Team SJ_AJ)

Language	Rank	Precision	Recall	F1
Kannada	1	0.73	0.78	0.75
Malayalam	2	0.97	0.97	0.96
Tamil	3	0.75	0.79	0.76

Data & Models

• See ./code/datasets/eacl2021 for creating data and running code. However, the main scripts are located in code/scripts folder.
• The .ipynb files (colab files) use absolute paths. So be careful and please change them according to your local directory path(s).
• See ./code/datasets/eacl2021/run.sh for how to run the code. Alternatively, see notebooks in ./colab_notebooks

Pretrained Models

• Pretrained models, submission files and training checkpoints can be downloaded from this drive repo.
• Scripts for task-adaptive pretraining are placed at ./pretraining

Citation

@misc{jayanthi2021sjajdravidianlangtecheacl2021,
      title={SJ_AJ@DravidianLangTech-EACL2021: Task-Adaptive Pre-Training of Multilingual BERT models for Offensive Language Identification}, 
      author={Sai Muralidhar Jayanthi and Akshat Gupta},
      year={2021},
      eprint={2102.01051},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Contact

Feel free to contact us for a quick chat or a discussion at gmail