TRAM: A Tripartite Multi-Agent Framework with Retrieval-Augmented Generation for Multi-Emotional Comment Classification--TRAM for Toxic Content Classification
Abstract
Multi-Emotional Comment Classification is a critical sub-task in sentiment analysis, essential for robust toxic content detection. Traditional generative agents relying solely on pre-trained Large Language Models (LLMs) often struggle with inherent biases, stereotypical outputs, and zero-shot hallucinations. To address these limitations, we propose a novel tripartite multi-agent framework that integrates Retrieval-Augmented Generation (RAG) to ground LLM reasoning in verified, domain-specific semantic context. Furthermore, to ensure rigorous and fair evaluation, we introduce a re-engineered and rebalanced subset of the BeaverTails dataset, mitigating the long-tail distribution bias common in toxicity benchmarks. We conducted a comprehensive ablation study comparing various RAG strategies (Naive, Re-ranking, Multi-Query) and Chain-of-Thought (CoT) reasoning across DeepSeek-V3.2 and LLaMA-3.3-70B-Instruct models. Empirical results demonstrate that our framework significantly enhances classification performance over standalone models. Notably, the synergistic combination of Multi-Query RAG and CoT reasoning on the LLaMA-3.3-70B architecture achieved a global optimum, recording a peak accuracy of 0.7139 and an F1-score of 0.7160. This study validates that external semantic grounding, combined with functionally decoupled multi-agent architecture, provides a robust, deterministic, and highly explainable blueprint for safety-critical toxic content moderation.
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.