Detecting Unintended Social Bias in Toxic Language Datasets

TL;DR

This paper introduces ToxicBias, a new dataset for detecting social biases in toxic language datasets, and evaluates transformer models for bias identification and mitigation, highlighting the importance of addressing unintended social biases in NLP.

Contribution

The paper presents a curated dataset ToxicBias with annotated social bias categories and provides baseline transformer model performance for bias detection and analysis.

Findings

Transformer models can identify social biases in toxic language datasets.

Bias mitigation strategies are discussed and evaluated.

The dataset enables systematic bias analysis in toxic language data.

Abstract

With the rise of online hate speech, automatic detection of Hate Speech, Offensive texts as a natural language processing task is getting popular. However, very little research has been done to detect unintended social bias from these toxic language datasets. This paper introduces a new dataset ToxicBias curated from the existing dataset of Kaggle competition named "Jigsaw Unintended Bias in Toxicity Classification". We aim to detect social biases, their categories, and targeted groups. The dataset contains instances annotated for five different bias categories, viz., gender, race/ethnicity, religion, political, and LGBTQ. We train transformer-based models using our curated datasets and report baseline performance for bias identification, target generation, and bias implications. Model biases and their mitigation are also discussed in detail. Our study motivates a systematic extraction of social bias data from toxic language datasets. All the codes and dataset used for experiments in this work are publicly available