Towards Understanding Gender Bias in Relation Extraction

TL;DR

This paper introduces WikiGenderBias, a dataset for evaluating gender bias in neural relation extraction systems, revealing existing biases and analyzing mitigation strategies like anonymization and debiasing.

Contribution

It creates the first dedicated dataset for gender bias analysis in NRE and evaluates how various techniques impact bias and system performance.

Findings

NRE systems exhibit significant gender bias in predictions.

Name anonymization and debiasing methods can reduce bias.

Bias mitigation techniques may affect overall system accuracy.

Abstract

Recent developments in Neural Relation Extraction (NRE) have made significant strides towards Automated Knowledge Base Construction (AKBC). While much attention has been dedicated towards improvements in accuracy, there have been no attempts in the literature to our knowledge to evaluate social biases in NRE systems. We create WikiGenderBias, a distantly supervised dataset with a human annotated test set. WikiGenderBias has sentences specifically curated to analyze gender bias in relation extraction systems. We use WikiGenderBias to evaluate systems for bias and find that NRE systems exhibit gender biased predictions and lay groundwork for future evaluation of bias in NRE. We also analyze how name anonymization, hard debiasing for word embeddings, and counterfactual data augmentation affect gender bias in predictions and performance.