TREC-IS 2020A, 2020B and 2021A".
Source code for "UCD participation in*** update at: 2021/05/25
This repo so far relates to the following work:
- Transformer-based Multi-task Learning for Disaster Tweet Categorisation, (WiP paper, ISCRAM 2021)
- Multi-task transfer learning for finding actionable information from crisis-related messages on social media, (paper, TREC 2020)
Setup
git clone https://github.com/wangcongcong123/crisis-mtl.git
pip install -r requirements.txt
Dataset preparation
- Download the splits prepared for the system from here that contains three subdirectories for 2020a, 2020b and 2021a respectively.
- Unzip the file to
data/
.
Training and submitting
# for 2020a
python run.py --dataset_name 2020a --model_name bert-base-uncased
# or for 2020b
python run.py --edition 2020b --model_name bert-base-uncased
python run.py --edition 2020b --model_name google/electra-base-discriminator
python run.py --edition 2020b --model_name microsoft/deberta-base
python run.py --edition 2020b --model_name distilbert-base-uncased
python submit_ensemble.py --edition 2020b
# or for 2021a
python run.py --edition 2021a --model_name bert-base-uncased
python run.py --edition 2021a --model_name google/electra-base-discriminator
python run.py --edition 2021a --model_name microsoft/deberta-base
python run.py --edition 2021a --model_name distilbert-base-uncased
python submit_ensemble.py --edition 2021a
To see our results compared to other participating runs in 2020a and 2020b, check the appendix of this overview paper. To know the details of our approach, check this ISCRAM-2021 paper on 2020a and this TREC-2020 paper on 2020b. The evaluation for 2021a is still in process so the results will be added as soon as they come out.
Citation
If you use the code in your research, please consider citing the following papers:
@article{wang2021,
author = {Wang, Congcong and Nulty, Paul and Lillis, David},
journal = {Proceedings of the International ISCRAM Conference},
keywords = {18th International Conference on Information Systems for Crisis Response and Management (ISCRAM 2021)},
number = {May},
title = {{Transformer-based Multi-task Learning for Disaster Tweet Categorisation}},
volume = {2021-May},
year = {2021}
}
@inproceedings{congcong2020multi,
address = {Gaithersburg, MD},
title = {Multi-task transfer learning for finding actionable information from crisis-related messages on social media},
booktitle = {Proceedings of the Twenty-Nineth {{Text REtrieval Conference}} ({{TREC}} 2020)},
author = {Wang, Congcong and Lillis, David},
year = {2020},
}
Queries
Let me know if any questions via [email protected] or through creating an issue.