ConvApparel: A Benchmark Dataset and Validation Framework for User Simulators in Conversational Recommenders

TL;DR

This paper introduces ConvApparel, a new dataset and validation framework for user simulators in conversational recommenders, addressing the realism gap and improving simulator robustness through counterfactual validation.

Contribution

It presents a novel dataset with dual-agent data collection and a comprehensive validation framework to evaluate and enhance user simulators in conversational AI.

Findings

Data-driven simulators outperform prompted baselines in realism.

Simulators adapt more realistically to unseen behaviors.

Significant realism gap identified across all tested simulators.

Abstract

The promise of LLM-based user simulators to improve conversational AI is hindered by a critical "realism gap," leading to systems that are optimized for simulated interactions, but may fail to perform well in the real world. We introduce ConvApparel, a new dataset of human-AI conversations designed to address this gap. Its unique dual-agent data collection protocol -- using both "good" and "bad" recommenders -- enables counterfactual validation by capturing a wide spectrum of user experiences, enriched with first-person annotations of user satisfaction. We propose a comprehensive validation framework that combines statistical alignment, a human-likeness score, and counterfactual validation to test for generalization. Our experiments reveal a significant realism gap across all simulators. However, the framework also shows that data-driven simulators outperform a prompted baseline, particularly in counterfactual validation where they adapt more realistically to unseen behaviors, suggesting they embody more robust, if imperfect, user models.