Louis-Alexandre Dit Petit-Frere, J., & Waldner, M. (2023). Visual Exploration of Indirect Bias in Language Models. In T. Hoelt, W. Aigner, & B. Wang (Eds.), EuroVis 2023 - Short Papers. The Eurographics Association. https://doi.org/10.2312/evs.20231034
E193-02 - Forschungsbereich Computer Graphics E193 - Institut für Visual Computing and Human-Centered Technology
-
Published in:
EuroVis 2023 - Short Papers
-
ISBN:
978-3-03868-219-6
-
Date (published):
2023
-
Event name:
25th EG Conference on Visualization (EuroVis 2023)
en
Event date:
12-Jun-2023 - 16-Jun-2023
-
Event place:
Leipzig, Germany
-
Number of Pages:
5
-
Publisher:
The Eurographics Association
-
Keywords:
visual exploration; language models; bias
en
Abstract:
Language models are trained on large text corpora that often include stereotypes. This can lead to direct or indirect bias in downstream applications. In this work, we present a method for interactive visual exploration of indirect multiclass bias learned by contextual word embeddings. We introduce a new indirect bias quantification score and present two interactive visualizations to explore interactions between multiple non-sensitive concepts (such as sports, occupations, and beverages) and sensitive attributes (such as gender or year of birth) based on this score.
en
Project title:
Joint Human-Machine Data Exploration: P 36453-N (FWF Fonds zur Förderung der wissenschaftlichen Forschung (FWF))
CC BY 4.0
