N²bias: Debiasing Long Inputs One Span at a Time

• Berkeley MIDS DATASCI 266 Natural Language Processing
• Tara Verma, Dave Russell

Files

• bias_detect.ipynb
  • Where we trained our models, both the NER model and our baseline binary classifer
• error_labeling.ipynb
  • Where we evaluated the performance of our NER model, scoring the classified token outputs
• baseline_dbias.ipynb
  • We considered a 2021 classifer, Dbias, as a baseline. The classifier is published by the same authors of the paper and dataset that inspired our work. However, it is a bit stale and the performance did not meet our expectations.
• nbias_eda.ipynb
  • Where we conducted an exploratory data analysis of the Nbias dataset.
• N²Bias Report
• Presentation Slides