Research

Paper

TESTING February 18, 2026

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

Authors

Ofer Meshi, Krisztian Balog, Sally Goldman, Avi Caciularu, Guy Tennenholtz, Jihwan Jeong, Amir Globerson, Craig Boutilier

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.

Metadata

arXiv ID: 2602.16938
Provider: ARXIV
Primary Category: cs.CL
Published: 2026-02-18
Fetched: 2026-02-21 18:51

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.16938v1</id>\n    <title>ConvApparel: A Benchmark Dataset and Validation Framework for User Simulators in Conversational Recommenders</title>\n    <updated>2026-02-18T23:00:21Z</updated>\n    <link href='https://arxiv.org/abs/2602.16938v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.16938v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>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.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-02-18T23:00:21Z</published>\n    <arxiv:comment>EACL 2026</arxiv:comment>\n    <arxiv:primary_category term='cs.CL'/>\n    <author>\n      <name>Ofer Meshi</name>\n    </author>\n    <author>\n      <name>Krisztian Balog</name>\n    </author>\n    <author>\n      <name>Sally Goldman</name>\n    </author>\n    <author>\n      <name>Avi Caciularu</name>\n    </author>\n    <author>\n      <name>Guy Tennenholtz</name>\n    </author>\n    <author>\n      <name>Jihwan Jeong</name>\n    </author>\n    <author>\n      <name>Amir Globerson</name>\n    </author>\n    <author>\n      <name>Craig Boutilier</name>\n    </author>\n  </entry>"
}