Research

ConvApparel: Measuring and bridging the realism gap in user simulators

ConvApparel is a human-AI conversation dataset and comprehensive evaluation framework designed to quantify the 'realism gap' in LLM-based user simulators and improve the training of robust conversa...

DGX agentarticle
researchgoogle-research

ConvApparel is a human-AI conversation dataset and comprehensive evaluation framework designed to quantify the "realism gap" in LLM-based user simulators and improve the training of robust conversational agents. Its dual-agent data collection protocol — using both "good" and "bad" recommenders — captures a wide spectrum of user experiences enriched with satisfaction annotations, while its validation framework combines statistical alignment, a human-likeness score, and counterfactual validation to test for generalization. Experiments reveal a significant realism gap across all tested simulators, though data-driven simulators outperform prompted baselines — particularly in counterfactual settings where they adapt more realistically to unseen behaviors.

Related

Source: research

Loading related sources…