STILE: Exploring and Debugging Social Biases in Pre-trained Text Representations

要旨

The recent success of Natural Language Processing (NLP) relies heavily on pre-trained text representations such as word embeddings. However, pre-trained text representations may exhibit social biases and stereotypes, e.g., disproportionately associating gender with occupations. Though prior work presented various bias detection algorithms, they are limited to pre-defined biases and lack effective interaction support. In this work, we propose STILE, an interactive system that supports mixed-initiative bias discovery and debugging in pre-trained text representations. STILE provides users the flexibility to interactively define and customize biases to detect based on their interests. Furthermore, it provides a bird’s-eye view of detected biases in a Chord diagram and allows users to dive into the training data to investigate how a bias was developed. Our lab study and expert review confirm the usefulness and usability of STILE as an effective aid in identifying and understanding biases in pre-trained text representations.

著者
Samia Kabir
Purdue University, West Lafayette, Indiana, United States
Lixiang Li
Purdue University, West Lafayette, Indiana, United States
Tianyi Zhang
Purdue University, West Lafayette, Indiana, United States
論文URL

https://doi.org/10.1145/3613904.3642111

動画

会議: CHI 2024

The ACM CHI Conference on Human Factors in Computing Systems (https://chi2024.acm.org/)

セッション: Ethics of Digital Technologies A

319
5 件の発表
2024-05-15 23:00:00
2024-05-16 00:20:00