EIBench: A Simulator-Based Benchmark and Turn-Credit RL for Emotion Management

2026-06-13

AI Summary

Current large language models struggle to capture the dynamic evolution of user emotions and relational states in multi-turn interactions. To address this limitation, this work introduces EIBench—the first emotion management evaluation benchmark that supports complex social behaviors such as boundary maintenance. EIBench comprises 2,222 dialogue trajectories spanning four scenarios: support, defense, repair, and charm, with an LLM-driven simulator tracking fine-grained emotion–relationship state transitions to provide turn-level dense feedback and final rewards. Building on this benchmark, we propose Centered Turn-Credit GRPO (CTC-GRPO), a novel algorithm that uniquely incorporates per-turn state updates as dense signals into reinforcement learning, jointly optimizing both interaction process and outcome. Experiments show that CTC-GRPO improves Qwen3-8B’s EIBench score from −22.4 to +22.4 and achieves gains of 12.4% and 20.9% on SAGE and EQBench3, respectively, substantially enhancing the model’s multi-turn emotional intelligence.

Abstract

Emotional intelligence (EI) in Large Language Models (LLMs) is often evaluated through static understanding tasks or single-response dialogue generation. However, emotion management is interactive: a good model should not only recognize a user's emotion, but also improve the user's emotional and relational state over several turns. We introduce EIBench, a simulator-based benchmark for interactive emotion management. EIBench contains 2,222 scenarios, with 2,009 for training and 213 for held-out testing. The scenarios are organized by a 2x2 taxonomy covering Support, Defense, Repair, and Charm, which together capture different forms of support, boundary maintenance, trust repair, and rapport building. In each scenario, an LLM simulator plays the user, updates an emotion-relation state after each turn, and maps the final state to an anchor-based score. This design makes EIBench both an evaluation benchmark and a training environment: the final state gives the outcome reward, while the per-turn state updates provide dense feedback for RL. We evaluate 15 open- and closed-source LLMs. Current models perform well on support and rapport-building scenes, but struggle with boundary maintenance under user pressure. To improve the EI ability of LLMs, we propose Centered Turn-Credit GRPO (CTC-GRPO), a GRPO extension that reuses the simulator's per-turn state updates as dense turn-level feedback while preserving the final outcome reward. CTC-GRPO improves Qwen3-8B from -22.4 to +22.4 on EIBench and also improves on out-of-distribution evaluations including SAGE (+12.4) and EQBench3 (+20.9%). Our results show that simulator-tracked user states can support both evaluation and training for multi-turn emotion management.

Problem

• Emotional Intelligence
• Large Language Models
• Interactive Emotion Management
• Multi-turn Dialogue
• Emotion-Relation State

Innovation

• EIBench
• emotion management
• simulator-based benchmark
• turn-credit reinforcement learning
• CTC-GRPO