Closing the Loop: Testing ChatGPT to Generate Model Explanations to Improve Human Labelling of Sponsored Content on Social Media

Thales Bertaglia, Stefan Huber, Catalina Goanta, Gerasimos Spanakis, Adriana Iamnitchi

Paper link

Data

The file dataset_labels.csv contains all posts used in the annotation task. Each row includes a post (represented by its shortcode) and each column is the label given by the corresponding annotator. See Section 3.4 of the paper to understand what each subgroup represents. You can access the posts on Instagram through instagram.com/p/{shortcode}.

The file prompt.txt contains the final version of the ChatGPT prompt used to generate the explanations used to augment the annotation task.

Citation

@misc{bertaglia2023closing,
      title={Closing the Loop: Testing ChatGPT to Generate Model Explanations to Improve Human Labelling of Sponsored Content on Social Media}, 
      author={Thales Bertaglia and Stefan Huber and Catalina Goanta and Gerasimos Spanakis and Adriana Iamnitchi},
      year={2023},
      eprint={2306.05115},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}