This repo contains code related to our submission for the task of Profiling Fake News Spreaders in PAN at CLEF 2020. The software directory include the scripts used in TIRA, verbatim. PyTorch model weights are required to run the scripts. Run these scripts the same as they would run in TIRA as:
./electra_{en/es}_{type}.py -i $inputDataset -o $outputDir
It expects data to be in XML format as specified by the organizers in the task homepage. You will need to edit the shebang to match your environment.
The model weights and Ensemble Training Notebooks can be viewed/downloaded from Kaggle:
EDA of the Dataset: EDA Notebook
Analysis of Ensemble: Notebook
Training data can be requested from Zenodo.
The desciption of the scripts are as follows:
electra_{en/es}_ensemble.py: Runs the complete ensemble of 15 models on the given inputDataset
./electra_{en/es}__ensemble.py -i inputDatasetDir -o outputDir -m savedModelsDir
electra_{en/es}_oneshot.py: Runs the best model found during once on the given inputDataset only once.
./electra_{en/es}__oneshot.py -i inputDatasetDir -o outputDir -m bestmodel.pt
electra_{en/es}_solo.py: Creates the ensemble using 15 copies of the best model and runs it on the given inputDataset.
./electra_{en/es}__solo.py -i inputDatasetDir -o outputDir -m bestmodel.pt
Requirements:
- This work reuses code written for another project which must be pulled as well
git clone --recurse-submodules https://github.com/cozek/trac2020_submission.git - Other libraries:
- PyTorch
- Transformers
- Pandas
- Numpy
- Scikit-learn
If code/paper was helpful, please cite:
@InProceedings{das:2020,
author = {{Kaushik Amar} Das and Arup Baruah and {Ferdous Ahmed} Barbhuiya and Kuntal Dey},
booktitle = {{CLEF 2020 Labs and Workshops, Notebook Papers}},
crossref = {pan:2020},
editor = {Linda Cappellato and Carsten Eickhoff and Nicola Ferro and Aur{\'e}lie N{\'e}v{\'e}ol},
month = sep,
publisher = {CEUR-WS.org},
title = {{Ensemble of ELECTRA for Profiling Fake News Spreaders---Notebook for PAN at CLEF 2020}},
url = {},
year = 2020
}