Debiasing knowledge graph embeddings

Joseph Fisher, Arpit Mittal, Dave Palfrey, Christos Christodoulopoulos


Abstract
It has been shown that knowledge graph embeddings encode potentially harmful social biases, such as the information that women are more likely to be nurses, and men more likely to be bankers. As graph embeddings begin to be used more widely in NLP pipelines, there is a need to develop training methods which remove such biases. Previous approaches to this problem both significantly increase the training time, by a factor of eight or more, and decrease the accuracy of the model substantially. We present a novel approach, in which all embeddings are trained to be neutral to sensitive attributes such as gender by default using an adversarial loss. We then add sensitive attributes back on in whitelisted cases. Training time only marginally increases over a baseline model, and the debiased embeddings perform almost as accurately in the triple prediction task as their non-debiased counterparts.
Anthology ID:
2020.emnlp-main.595
Volume:
Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)
Month:
November
Year:
2020
Address:
Online
Editors:
Bonnie Webber, Trevor Cohn, Yulan He, Yang Liu
Venue:
EMNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
7332–7345
Language:
URL:
https://aclanthology.org/2020.emnlp-main.595
DOI:
10.18653/v1/2020.emnlp-main.595
Bibkey:
Cite (ACL):
Joseph Fisher, Arpit Mittal, Dave Palfrey, and Christos Christodoulopoulos. 2020. Debiasing knowledge graph embeddings. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 7332–7345, Online. Association for Computational Linguistics.
Cite (Informal):
Debiasing knowledge graph embeddings (Fisher et al., EMNLP 2020)
Copy Citation:
PDF:
https://aclanthology.org/2020.emnlp-main.595.pdf
Video:
 https://slideslive.com/38938899