The Young Researcher (Aug 2024)
Understanding how ChatGPT’s political bias impacts hate speech and offensive language detection: A content analysis.
Abstract
Given the popularity and demonstrated proficiency of ChatGPT in detecting hate speech, it is important to understand the potential influence of its left-libertarian bias in potential tasks. This study examines how ChatGPT’s political bias affects its detection of hate speech and offensive language in X posts specifically focused on two partisan issues: gun control and abortion rights. The research finds a minor left-leaning bias in hate speech detection, particularly evident for the abortion rights issue. The findings suggest that while ChatGPT’s bias could influence content moderation decisions, its magnitude is relatively minor. ChatGPT had an accuracy rate of 68% for left-leaning posts and 59% for right-leaning posts.
