Data Feminism Bias Report Generator

Introduction

This is a prototype tool for authors of NLP research papers to report biases related to their datasets. It dynamically adapts to information about the dataset as you add it to suggest possible biases. Each bias includes a citation to published statistics and previous research.

If there is a research paper you think should be added to the tool, you can suggested it through this form.

All code for the prototype is published on our GitHub.

You can also read the paper (forthcoming) to understand the research upon which this prototype is based.

Inspiration

There are other bias reporting paradigms that we drew inspiration from in creating this tool. Often they are tailored to other types of datasets or models, or reporting contexts. Additionally, most still rely on paper authors to generate potential sources of bias, whereas our tool proactively suggests them. See the GitHub for a complete list.

Only one item is selectable. If your data come from more than one source, we recommend filling out multiple forms.

Estimated
Estimated
Estimated
Estimated