Research map / Multi-agent systems and games / LLM agents
Consensus-Driven Multi-Agent Cognitive Reasoning for Enhancing the Emotional Intelligence of Large Language Models
Geng Tu, Dingming Li, Jun Huang, Ruifeng Xu
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes MACRo, a consensus-driven multi-agent cognitive reasoning framework to improve the emotional intelligence of LLMs on EmoBench-style tasks.
Problem
LLMs remain limited in emotional intelligence, particularly on cognitively demanding tasks requiring inference of others' latent mental states, intentions, and emotions in nuanced social contexts (as benchmarked by EmoBench).
Approach
A multi-agent cognitive reasoning framework (MACRo) in which multiple agents reason about social-emotional scenarios and reach consensus to improve EI reasoning; details beyond the abstract are unavailable.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add MACRo CogCoT prompt | Baichuan2-53B · EmoBench EU overall (English) | accuracy (%): 34.88 -> 47.62 | Improved |
| add MACRo CogCoT prompt | DeepSeek-V3 · EmoBench EU overall (English) | accuracy (%): 57.50 -> 58.62 | Improved |
| add MACRo CogCoT prompt | Baichuan2-Chat-7B · EmoBench EU overall (English) | accuracy (%): 22.38 -> 14.62 | Hurt |
| add MACRo CogCoT prompt | Baichuan2-Chat-13B · EmoBench EU overall (English) | accuracy (%): 26.25 -> 22.50 | Hurt |
| add MACRo CogCoT prompt | Llama2-Chat-13B · EmoBench EU overall (English) | accuracy (%): 17.38 -> 16.00 | No clear effect |
| add MACRo CogCoT prompt | Qwen-Chat-14B · EmoBench EA overall (English) | accuracy (%): 60.50 -> 58.75 | Hurt |
| add MACRo CogCoT prompt | Baichuan2-53B · EmoBench EA overall (English) | accuracy (%): 65.38 -> 74.25 | Improved |
| add MACRo CogCoT prompt (general) | 11 LLMs · EmoBench EU/EA (EN, ZH) | accuracy (%): Baichuan2-Chat-7B EU EN 22.38 -> 14.62; Baichuan2-53B EU EN 34.88 -> 47.62 | Mixed |
| add consensus game (ablation) | MACRo (ChatGLM3-6B) · EmoBench EU overall (English) | accuracy (%): 22.75 -> 24.50 | Improved |
| add consensus game (ablation) | MACRo (DeepSeek-V3) · EmoBench EU overall (English) | accuracy (%): 57.75 -> 58.62 | Improved |
| add multi-agent role specialization (ablation) | MACRo (ChatGLM3-6B) · EmoBench EU overall (English) | accuracy (%): 20.38 -> 24.50 | Improved |
| use zero-shot MACRo CogCoT instead of trained COLM chains | Llama2-Chat-7B · EmoBench EA overall (Chinese) | accuracy (%): 31.75 -> 44.75 | Improved |
Experimental setup
Datasets: EmoBench
Benchmarks: EmoBench
Builds on: EmoBench
Closest papers by meaning
- MetaMind: Modeling Human Social Thoughts with Metacognitive Multi-Agent Systems (NeurIPS 2025 · similarity 0.89)
- Reflection-Bench: Evaluating Epistemic Agency in Large Language Models (ICML 2025 · similarity 0.88)
- Unlocking the Power of Multi-Agent LLM for Reasoning: From Lazy Agents to Deliberation (ICLR 2026 · similarity 0.88)
- Unveiling the Cognitive Compass: Theory-of-Mind–Guided Multimodal Emotion Reasoning (ICLR 2026 · similarity 0.88)
- Systematic Failures in Collective Reasoning under Distributed Information in Multi-Agent LLMs (ICML 2026 · similarity 0.87)
- Eliciting Reasoning in Language Models with Cognitive Tools (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
