Reducing Target Group Bias in Hate Speech Detectors

TL;DR

This paper identifies biases in hate speech detection models across different target groups and proposes token-level hate sense disambiguation to improve fairness and accuracy for underrepresented groups.

Contribution

It introduces a token-level hate sense disambiguation method that reduces target group bias and enhances overall fairness in hate speech detection models.

Findings

Accuracy on underrepresented groups improved by up to 13%.

Variance in model accuracy across groups decreased by at least 30%.

Average performance increased by 4% across datasets.

Abstract

The ubiquity of offensive and hateful content on online fora necessitates the need for automatic solutions that detect such content competently across target groups. In this paper we show that text classification models trained on large publicly available datasets despite having a high overall performance, may significantly under-perform on several protected groups. On the \citet{vidgen2020learning} dataset, we find the accuracy to be 37\% lower on an under annotated Black Women target group and 12\% lower on Immigrants, where hate speech involves a distinct style. To address this, we propose to perform token-level hate sense disambiguation, and utilize tokens' hate sense representations for detection, modeling more general signals. On two publicly available datasets, we observe that the variance in model accuracy across target groups drops by at least 30\%, improving the average target group performance by 4\% and worst case performance by 13\%.