This is a NLP project examining BERT-based abstractive summarization model for gender bias. We showed that change in word-choice sentiment from input news articles to output summaries is slightly more negative for articles exclusively about women than for articles exclusively about men.
The code should run without modifications. Just download the repository and follow along with the notebooks if you would like to experiment with the data or reproduce the results in our report.
To see the final report, click this link or the thumbnail, or navigate to final_report.pdf
.