CC-Mediation: Evaluating Large Language Models for Cross-Cultural Conflict Mediation

The research team introduced the CC-Mediation benchmark, aimed at addressing the lack of measurable datasets and principled evaluation metrics in cross-cultural conflict mediation. The benchmark includes 1,661 rounds of dialogue data based on the DMIS model, and proposes two DMIS-based evaluation metrics: Trajectory AUC is used to measure the持续性 of cross-cultural improvement, while the signed Wasserstein-1 distance is used to measure the magnitude and direction of position shifts. Experiments revealed that current large language models have limitations in both timing of intervention and mediation strategies: the former stems from ignoring the hierarchical prior of dialogue content, while the latter results from the collapse of deep networks rather than knowledge deficiency.