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

EIBench is introduced as a simulator-based benchmark designed for evaluating and training Large Language Models (LLMs) in interactive emotion management, comprising 2,222 scenarios divided into training and testing sets. The benchmark employs a 2x2 taxonomy covering various emotional support strategies and utilizes a turn-based feedback mechanism to enhance model performance through a novel reinforcement learning approach called Centered Turn-Credit GRPO (CTC-GRPO), which significantly improved the Qwen3-8B model's performance on EIBench and other evaluation metrics. This development is crucial for practitioners aiming to enhance LLMs' emotional intelligence capabilities in multi-turn dialogues.