dkorenci / gpt-def-zeroshot

Code and data for the experiments of the article

'Definitions Matter: Guiding GPT for Multi-label Classification'

CODE

The semantic understanding experiments are in the experiments/sem_understand_corr.py module

DATA

The experiment-generated data is in the 'data' folder: generated definitions, zero-shot classification results.

We also provide the labels and the IDs of the tweets. The IDs correspond to the test set used for the MediaEval 2022 FakeNews challenge.

This data is a subset of the COCO dataset from the paper: 'COCO: an annotated Twitter dataset of COVID-19 conspiracy theories' https://doi.org/10.1007/s42001-023-00200-3

Instructions how to obtain this datasets, and link to the script to download the original tweets can be found in the 'Availability of data and materials' paragraph of the COCO paper.