Re-ranking Using Large Language Models for Mitigating Exposure to Harmful Content on Social Media Platforms

2025-01-23

AI Summary

Social media recommendation algorithms often inadvertently expose users to harmful content. Method: This paper proposes a zero-shot/few-shot dynamic re-ranking framework leveraging large language models (LLMs), eliminating reliance on large-scale human-annotated data. It introduces a novel exposure-aware re-ranking evaluation metric specifically designed for mitigating harmful content exposure—bypassing traditional classifier dependencies on labeled data and fixed thresholds—to enhance scalability and real-time adaptability. Results: Evaluated across three public datasets, three baseline recommendation models, and multiple configurations, the approach significantly reduces harmful content exposure rates compared to mainstream commercial moderation systems, while preserving recommendation relevance and ranking quality. Key contributions include: (1) the first exposure-aware re-ranking paradigm; (2) a zero-shot-driven dynamic intervention mechanism; and (3) a generalizable safety–utility co-optimization framework that jointly balances content safety and recommendation effectiveness.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Data Mining & Knowledge Management: Recommender Systems
• Natural Language Processing: Safety and Robustness

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Social media platforms utilize Machine Learning (ML) and Artificial Intelligence (AI) powered recommendation algorithms to maximize user engagement, which can result in inadvertent exposure to harmful content. Current moderation efforts, reliant on classifiers trained with extensive human-annotated data, struggle with scalability and adapting to new forms of harm. To address these challenges, we propose a novel re-ranking approach using Large Language Models (LLMs) in zero-shot and few-shot settings. Our method dynamically assesses and re-ranks content sequences, effectively mitigating harmful content exposure without requiring extensive labeled data. Alongside traditional ranking metrics, we also introduce two new metrics to evaluate the effectiveness of re-ranking in reducing exposure to harmful content. Through experiments on three datasets, three models and across three configurations, we demonstrate that our LLM-based approach significantly outperforms existing proprietary moderation approaches, offering a scalable and adaptable solution for harm mitigation.

Problem

• Social Media Platforms
• Inappropriate Content
• Algorithmic Moderation

Innovation

• Advanced Language Models
• Information Ranking Optimization
• Performance in Novel Issues