About The REPO

This repoistory is a collaborative effort by the DCU - ADAPT Team for the EDOS 2023 SemEval Task-10 shared task.

Folder Structure

1. src/ -> scripts for all three tasks with transformer models
2. soc_sci_exp/ -> scripts and excel sheets with responses from 3 social science expert annotators and Chat-GPT and FLAN-T5 (large)
3. scripts -> scripts for experiments with baseline system and augmentation of training data

Pre-requisites

Python3 modules: gensim nltk numpy pandas scikit-learn spacy xgboost transformers sentence-transformers torch

(SpaCy is only needed if requesting its tokeniser with train.py --tokeniser, or making predictions with a model that was trained with this option.)

NLTK asks to further run inside python:

import nltk
nltk.download('punkt')

Spacy instructions say to run

python -m spacy download en_core_web_sm

TODO: do we need this model? We only use the tokeniser.

Usage

See task specific README.md in src/ directory

License

See LICENSE.txt

Citation

If you use this code or the released models, please cite our paper to appear at SemEval 2023.

Acknowledgements