Mod-Guide: An LLM-based Content Moderation Feedback System to Address Insensitive Speech toward Indigenous Ethnic and Religious Minority Communities

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Software Development & Engineering · Depth: Expert, extended

Summary

Mod-Guide, an LLM-based content moderation feedback system, addresses culturally insensitive speech targeting Bangladesh's Hindu and Chakma communities, the country's largest religious and Indigenous ethnic minorities. Developed between December 2024 and January 2025, this system integrates a co-created corpus of 132 culturally insensitive speech instances and community explanations into moderation pipelines using Retrieval Augmented Generation (RAG). A mixed-method evaluation, involving two experts and 15 participants, compared RAG-enhanced GPT-4 responses against off-the-shelf GPT-4. Results indicate that RAG significantly improves contextual accuracy and the perceived usefulness of moderation feedback, with usefulness varying across ethnic lines but not religious ones. This work contributes a unique dataset and the Mod-Guide artifact, advancing AI ethics and human-computer interaction by incorporating epistemically marginalized perspectives.

Key takeaway

For AI Ethicists and NLP Engineers designing content moderation systems for diverse online communities, recognize that standard LLMs often reinforce majority perspectives, leading to hermeneutical injustice. You should integrate Retrieval Augmented Generation (RAG) with community-sourced, culturally grounded corpora to ensure moderation reflects minority viewpoints. This approach, exemplified by Mod-Guide, significantly enhances contextual accuracy and perceived usefulness, fostering restorative justice and preventing epistemic erasure in your systems.

Key insights

LLMs can achieve culturally sensitive content moderation by integrating retrieval-augmented generation (RAG) with community-sourced, epistemically marginalized perspectives.

Principles

• Insensitive speech implicitly disregards minority cultural values.
• Content moderation needs restorative justice principles.
• Minority datasets are "prototype-based categories."

Method

Mod-Guide's method involves co-creating a culturally insensitive speech corpus via Asynchronous Remote Communities (ARC), integrating it into a GPT-4 RAG pipeline with persona prompting, and evaluating feedback through textual analysis, expert review, and user studies.

In practice

• Ground LLM moderation in community-sourced data via RAG.
• Use persona-based prompting for diverse moderation roles.
• Prioritize minority hermeneutics in data curation.

Topics

• LLM Content Moderation
• Retrieval-Augmented Generation
• AI Ethics
• Culturally Insensitive Speech
• Minority Perspectives
• Restorative Justice

Best for: Research Scientist, AI Scientist, AI Ethicist, NLP Engineer

Related on AIssential

• Mod-Guide: An LLM-based Content Moderation Feedback System to Address Insensitive Speech toward Indigenous Ethnic and Religious Minority Communities — Integrating culturally grounded narratives via RAG enhances LLM sensitivity to minority perspectives in content moderation.
• Retrieval-Augmented Generation Must Move Beyond Factual Grounding to Represent Diverse Opinions — Current RAG systems exhibit factual bias, necessitating Opinion-Aware RAG to represent diverse perspectives by preserving aleatoric uncertainty.
• More Context, Larger Models, or Moral Knowledge? A Systematic Study of Schwartz Value Detection in Political Texts — Value detection performance depends critically on context, external knowledge, and model architecture, not just scale.
• Can we stop these LLM posts and replies? [D] — AI-generated content is degrading online discourse, prompting calls for enhanced moderation and new detection strategies.
• When Cases Get Rare: A Retrieval Benchmark for Off-Guideline Clinical Question Answering — LLMs require retrieval augmentation and strong reasoning for reliable off-guideline medical question answering.
• Evaluating Pluralism in LLMs through Latent Perspectives — LLMs struggle to represent diverse perspectives, especially rarer ones, diverging from human text distributions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.