The Effects of Demographic Instructions on LLM Personas

TL;DR

This paper explores how incorporating demographic instructions into large language models enables personalized detection of sexist content on social media, addressing subjectivity and diversity in perceptions.

Contribution

It introduces a perspectivist approach that uses demographic data to personalize sexism detection, moving beyond standard labels to account for individual and group perspectives.

Findings

Personalized models better capture subjective views of sexism.

Demographic instructions improve detection accuracy for diverse user groups.

The approach enhances fairness and inclusivity in content moderation.

Abstract

Social media platforms must filter sexist content in compliance with governmental regulations. Current machine learning approaches can reliably detect sexism based on standardized definitions, but often neglect the subjective nature of sexist language and fail to consider individual users' perspectives. To address this gap, we adopt a perspectivist approach, retaining diverse annotations rather than enforcing gold-standard labels or their aggregations, allowing models to account for personal or group-specific views of sexism. Using demographic data from Twitter, we employ large language models (LLMs) to personalize the identification of sexism.