Pith. sign in

REVIEW 2 major objections 2 minor 36 references

AdaptSim uses automatic prompt generation and open actions to adapt user simulators across domains for reliable CRS evaluation.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-06-26 07:11 UTC pith:4VSFUBIX

load-bearing objection AdaptSim adds auto prompt tuning, open actions, and BFS pairwise comparison to LLM simulators for CRS, but the reliability of the evaluation still hinges on unshown controls for simulator artifacts. the 2 major comments →

arxiv 2606.22803 v2 pith:4VSFUBIX submitted 2026-06-22 cs.IR

Towards Fast Domain Adaptation and Fine-Grained User Simulation for Evaluating Conversational Recommender Systems

Yuanzi Li , Quanyu Dai , Xueyang Feng , Zihang Tian , Junhao Wang , Xu Chen , Zhenhua Dong , Huifeng Guo This is my paper
classification cs.IR
keywords conversational recommender systemsuser simulationdomain adaptationlarge language modelsevaluation frameworkprompt tuningdialogue generation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces AdaptSim to address key limits in LLM-based user simulators for conversational recommender systems. Fixed prompts and closed action spaces make those simulators hard to transfer to new domains and weak at capturing varied user styles. AdaptSim instead generates prompts automatically and uses an open action mechanism to cut manual tuning while supporting cross-domain use. Response generation follows a think-then-respond approach for style control, and evaluation runs through a BFS-based turn-level pairwise comparison framework. Experiments across three domains and four LLMs show the resulting dialogues support more effective and robust system testing.

Core claim

AdaptSim is an adaptive user simulator that employs automatic prompt generation and an open action mechanism to model realistic user behavior across domains, paired with a think-then-respond strategy for fine-grained style control and a BFS-based turn-level pairwise comparison framework for comprehensive CRS evaluation.

What carries the argument

AdaptSim's combination of automatic prompt generation, open action mechanism, think-then-respond response generation, and BFS-based turn-level pairwise comparison framework.

Load-bearing premise

Automatic prompt generation combined with an open action mechanism will produce realistic, unbiased user behavior that transfers to novel domains without manual tuning or evaluation-invalidating artifacts.

What would settle it

Human evaluators in a blind test rate AdaptSim dialogues as substantially less realistic than real user conversations, or the BFS framework ranks known strong CRSs below weaker ones across multiple runs.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • CRSs can be assessed for core capabilities and robustness using simulations that transfer across domains without per-domain redesign.
  • User modeling captures subtle linguistic styles and shifting preferences through controlled generation rather than fixed templates.
  • Evaluation moves beyond single-turn metrics to structured turn-level comparisons that expose interaction weaknesses.
  • The simulator reduces reliance on domain experts for prompt engineering when testing new recommendation settings.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same adaptation mechanism could support rapid prototyping of CRSs for emerging product categories where real user data is scarce.
  • Generated dialogues might serve as synthetic training data to improve the underlying recommender models themselves.
  • The BFS comparison structure could extend to evaluating other multi-turn dialogue systems such as task-oriented chatbots.
  • Combining the open action space with reinforcement learning might allow the simulator to evolve preferences over longer sessions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes AdaptSim, an adaptive user simulator for conversational recommender systems that uses automatic prompt generation and an open action mechanism to address limitations in domain adaptability, user modeling, and evaluation validity. It introduces a 'think-then-respond' strategy for controlled response generation and a novel BFS-based turn-level pairwise comparison framework for CRS evaluation. Experiments across three domains and four LLMs are presented to support claims of realistic dialogue generation and reliable assessment of CRS capabilities and robustness.

Significance. If the central claims on realism and lack of simulator artifacts hold, the work would offer a meaningful advance in CRS evaluation by reducing manual prompt and action-space engineering while enabling cross-domain transfer. The automatic prompt tuning and BFS framework represent potentially useful methodological contributions for scalable assessment.

major comments (2)
  1. [§5 (Experiments)] §5 (Experiments): The claim that AdaptSim 'generates realistic dialogues' enabling 'highly effective and reliable evaluation' lacks reported quantitative metrics for realism (e.g., human judgment scores, divergence from logged user actions), baseline comparisons with statistical significance, or ablation on post-hoc prompt tuning choices; without these, the central effectiveness claim cannot be assessed.
  2. [§4.3 (Open action mechanism)] §4.3 (Open action mechanism): The open action space is presented as removing bias from predefined constraints, yet no validation (e.g., comparison of action distributions to real-user logs or sensitivity analysis) rules out LLM-induced artifacts in action sequences; this is load-bearing for the BFS pairwise comparisons, as any systematic simulator bias would render cross-CRS differences uninterpretable.
minor comments (2)
  1. [Abstract] Abstract: The three limitations are listed but the mapping from each limitation to the corresponding AdaptSim component could be stated more explicitly for clarity.
  2. [§3 (Response generation)] §3 (Response generation): The 'think-then-respond' strategy is described at a high level; adding a short pseudocode snippet or example prompt template would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive feedback highlighting the need for stronger empirical validation of realism claims and the open action mechanism. We address each major comment below, agreeing where revisions are warranted while noting limitations on data availability.

read point-by-point responses

  1. Referee: [§5 (Experiments)] The claim that AdaptSim 'generates realistic dialogues' enabling 'highly effective and reliable evaluation' lacks reported quantitative metrics for realism (e.g., human judgment scores, divergence from logged user actions), baseline comparisons with statistical significance, or ablation on post-hoc prompt tuning choices; without these, the central effectiveness claim cannot be assessed.

    Authors: We acknowledge that the experiments in the current manuscript rely primarily on cross-domain and cross-LLM results to support effectiveness, without direct quantitative realism metrics such as human judgment scores or statistical significance tests against baselines. We will revise Section 5 to include human evaluation scores for dialogue realism, statistical tests for comparisons, and ablations on prompt tuning choices to better substantiate the claims. revision: yes

  2. Referee: [§4.3 (Open action mechanism)] The open action space is presented as removing bias from predefined constraints, yet no validation (e.g., comparison of action distributions to real-user logs or sensitivity analysis) rules out LLM-induced artifacts in action sequences; this is load-bearing for the BFS pairwise comparisons, as any systematic simulator bias would render cross-CRS differences uninterpretable.

    Authors: We agree that validation is essential for the open action mechanism given its role in the BFS framework. We will add sensitivity analysis on action sequence distributions in the revision to check for potential artifacts. Direct comparison to real-user logs is not feasible, as such logs are unavailable for the novel domains evaluated. revision: partial

standing simulated objections not resolved
  • Direct comparison of action distributions to real-user logs, as no such logs are available for the domains tested.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper introduces AdaptSim via automatic prompt generation, open action mechanism, controlled text generation with think-then-respond, and a BFS-based turn-level pairwise comparison for CRS evaluation. No equations, fitted parameters, or predictions are described that reduce by construction to inputs. The derivation chain relies on the proposed mechanisms and cross-domain experiments for validation rather than self-definition, fitted-input renaming, or load-bearing self-citations. The evaluation framework is presented as independent of simulator parameters, consistent with the reader's assessment of score 2.0 as the upper bound for minor issues.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only; no explicit free parameters, axioms, or invented entities are stated.

pith-pipeline@v0.9.1-grok · 5777 in / 1088 out tokens · 21332 ms · 2026-06-26T07:11:02.657324+00:00 · methodology

0 comments
read the original abstract

Conversational Recommender Systems (CRSs) enhance user experience through multi-turn interactions, yet evaluating their performance remains challenging. While Large Language Model (LLM) based user simulators are effective, they suffer from three key limitations: (1) Lack of Domain Adaptability: Reliance on fixed prompts and predefined action spaces hinders transfer to novel domains; (2) Limited User Modeling: Inability to accurately replicate subtle linguistic styles and dynamic preferences; (3) Insufficient Evaluation Validity: Existing simulators fail to adequately assess fundamental capabilities and system robustness. To overcome these, we propose AdaptSim, an Adaptive domain and automatic prompt tuning User Simulator. AdaptSim offers an efficient framework for evaluating CRSs by enabling realistic behavior modeling and diverse style generation. It leverages automatic prompt generation and an open action mechanism to reduce manual effort and improve cross-domain flexibility. For response generation, we employ controlled text generation with a "think-then-respond" strategy for fine-grained control over language style. For CRS evaluation, AdaptSim incorporates a novel Breadth-First Search (BFS)-based, turn-level pairwise comparison framework for comprehensive assessment. Extensive experiments across three domains and four LLMs demonstrate that AdaptSim generates realistic dialogues, enabling a highly effective and reliable evaluation of CRS capabilities and robustness.

Figures

Figures reproduced from arXiv: 2606.22803 by Huifeng Guo, Junhao Wang, Quanyu Dai, Xu Chen, Xueyang Feng, Yuanzi Li, Zhenhua Dong, Zihang Tian.

Figure 1
Figure 1. Figure 1: An illustration of the three core capabilities of an ideal user simulator. (a) [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) User Profile Shows the multi-domain profiles, demonstrating that our simulator can quickly adapt to new domains. (b) Prompt Optimization Illustrates the automatic prompt optimization process when adapt￾ing to a new domain: first, domain adaptation is performed, followed by iterative optimization of the prompt based on interactions with CRS. (c) Simulator. After optimizing the prompt, we integrate it in… view at source ↗
Figure 3
Figure 3. Figure 3: Pairwise win–draw–loss comparison of AdaptSim against three baseline simulators across the Food, [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Human validation of the LLM-as-a-judge protocol. Three senior annotators in the recommender sys [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The adherence of the user simulator to the designated style across four LLM backbones and four [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of AdaptSim vs. Prompting Methods in [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Ablation analysis of open-ended action generation. The figure compares an open action space with [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Pairwise evaluation of CRS performance under normal and careless user conditions. Here, BC denotes [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Pearson and Spearman correlation of AdaptSim evaluation scores across backbone LLM pairs and task [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The first bad case of RecuserSim based on GPT-4o backbone, where improper refinement led to role [PITH_FULL_IMAGE:figures/full_fig_p028_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: The second bad case of RecuserSim based on GPT-4o backbone, where improper refinement led to [PITH_FULL_IMAGE:figures/full_fig_p029_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: The first bad case of iEvaLM based on the GPT-4o backbone, where the lack of strategic guidance [PITH_FULL_IMAGE:figures/full_fig_p030_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: The second bad case of iEvaLM based on the GPT-4o backbone, where the lack of strategic guidance [PITH_FULL_IMAGE:figures/full_fig_p031_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: The first bad case of CSHI based on the GPT-4o backbone, where limited strategies led to repeated ACM TransInfSystVol1No1Article 1Publication date: January 2025 [PITH_FULL_IMAGE:figures/full_fig_p032_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: The second bad case of CSHI based on the GPT-4o backbone, where limited strategies led to repeated ACM Trans. Inf. Syst., Vol. 1, No. 1, Article 1. Publication date: January 2025. [PITH_FULL_IMAGE:figures/full_fig_p033_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: The good case of AdaptSim based on the GPT-4o backbone, where the fine-grained formal style is [PITH_FULL_IMAGE:figures/full_fig_p034_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: The good case of AdaptSim based on the GPT-4o backbone, where the fine-grained informal style is [PITH_FULL_IMAGE:figures/full_fig_p035_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: The good case of AdaptSim based on the GPT-4o backbone, where the fine-grained long style is [PITH_FULL_IMAGE:figures/full_fig_p036_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: The good case of AdaptSim based on the GPT-4o backbone, where the fine-grained short style is [PITH_FULL_IMAGE:figures/full_fig_p037_19.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

36 extracted references · 20 canonical work pages · 5 internal anchors

  1. [1]

    Yupeng Chang, Xu Wang, Jindong Wang, Yuan Wu, Linyi Yang, Kaijie Zhu, Hao Chen, Xiaoyuan Yi, Cunxiang Wang, Yidong Wang, et al. 2024. A survey on evaluation of large language models. ACM transactions on intelligent systems and technology 15, 3 (2024), 1–45

    2024

  2. [2]

    Luyu Chen, Quanyu Dai, Zeyu Zhang, Xueyang Feng, Mingyu Zhang, Pengcheng Tang, Xu Chen, Yue Zhu, and Zhen- hua Dong. 2025. RecUserSim: A Realistic and Diverse User Simulator for Evaluating Conversational Recommender Systems. In Companion Proceedings of the ACM on Web Conference 2025 . 133–142

    2025

  3. [3]

    Jiabao Fang, Shen Gao, Pengjie Ren, Xiuying Chen, Suzan Verberne, and Zhaochun Ren. 2024. A multi-agent conver- sational recommender system. arXiv preprint arXiv:2402.01135 (2024)

    work page arXiv 2024

  4. [4]

    Chongming Gao, Wenqiang Lei, Xiangnan He, Maarten De Rijke, and Tat-Seng Chua. 2021. Advances and challenges in conversational recommender systems: A survey. AI open 2 (2021), 100–126

    2021

  5. [6]

    Yunfan Gao, Tao Sheng, Youlin Xiang, Yun Xiong, Haofen Wang, and Jiawei Zhang. 2023. Chat-rec: Towards interac- tive and explainable llms-augmented recommender system. arXiv preprint arXiv:2303.14524 (2023)

    work page arXiv 2023

  6. [7]

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  7. [8]

    Zhankui He, Zhouhang Xie, Rahul Jha, Harald Steck, Dawen Liang, Yesu Feng, Bodhisattwa Prasad Majumder, Nathan Kallus, and Julian McAuley. 2023. Large language models as zero-shot conversational recommenders. In Proceedings of the 32nd ACM international conference on information and knowledge management . 720–730

    2023

  8. [9]

    Xu Huang, Jianxun Lian, Yuxuan Lei, Jing Yao, Defu Lian, and Xing Xie. 2023. Recommender ai agent: Integrating large language models for interactive recommendations. arXiv preprint arXiv:2308.16505 (2023)

    work page arXiv 2023

  9. [10]

    Rolf Jagerman, Ilya Markov, and Maarten de Rijke. 2019. When people change their mind: Off-policy evaluation in non-stationary recommendation environments. In Proceedings of the twelfth ACM international conference on web search and data mining . 447–455

    2019

  10. [11]

    Dietmar Jannach, Ahtsham Manzoor, Wanling Cai, and Li Chen. 2021. A survey on conversational recommender systems. ACM Computing Surveys (CSUR) 54, 5 (2021), 1–36

    2021

  11. [12]

    Dietmar Jannach, Ahtsham Manzoor, Wanling Cai, and Li Chen. 2021. A Survey on Conversational Recommender Systems. ACM Comput. Surv. 54, 5 (2021)

    2021

  12. [13]

    Knill and Alexandre Pouget

    David C. Knill and Alexandre Pouget. 2004. The Bayesian brain: the role of uncertainty in neural coding and compu- tation. Trends in Neurosciences 27, 12 (2004), 712–719. https://doi.org/10.1016/j.tins.2004.10.007

    work page doi:10.1016/j.tins.2004.10.007 2004

  13. [14]

    Wenqiang Lei, Gangyi Zhang, Xiangnan He, Yisong Miao, Xiang Wang, Liang Chen, and Tat-Seng Chua. 2020. Inter- active path reasoning on graph for conversational recommendation. In Proceedings of the 26th ACM SIGKDD interna- tional conference on knowledge discovery & data mining . 2073–2083

    2020

  14. [15]

    Lei Li, Yongfeng Zhang, Dugang Liu, and Li Chen. 2023. Large language models for generative recommendation: A survey and visionary discussions. arXiv preprint arXiv:2309.01157 (2023)

    work page arXiv 2023

  15. [16]

    Qingquan Li, Shaoyu Dou, Kailai Shao, Chao Chen, and Haixiang Hu. 2025. Evaluating Scoring Bias in LLM-as-a-Judge. arXiv preprint arXiv:2506.22316 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  16. [17]

    Xun Liang, Hanyu Wang, Yezhaohui Wang, Shichao Song, Jiawei Yang, Simin Niu, Jie Hu, Dan Liu, Shunyu Yao, Feiyu Xiong, and Zhiyu Li. 2024. Controllable Text Generation for Large Language Models: A Survey. arXiv:2408.12599 [cs.CL] https://arxiv.org/abs/2408.12599

    work page arXiv 2024

  17. [18]

    Lizi Liao, Ryuichi Takanobu, Yunshan Ma, Xun Yang, Minlie Huang, and Tat-Seng Chua. 2019. Deep conversational recommender in travel. arXiv preprint arXiv:1907.00710 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  18. [19]

    Yuanxing Liu, Wei-Nan Zhang, Yifan Chen, Yuchi Zhang, Haopeng Bai, Fan Feng, Hengbin Cui, Yongbin Li, and Wanxiang Che. 2023. Conversational recommender system and large language model are made for each other in E-commerce pre-sales dialogue. arXiv preprint arXiv:2310.14626 (2023)

    work page arXiv 2023

  19. [20]

    Reid Pryzant, Dan Iter, Jerry Li, Yin Tat Lee, Chenguang Zhu, and Michael Zeng. 2023. Automatic prompt optimization with” gradient descent” and beam search. arXiv preprint arXiv:2305.03495 (2023)

    work page arXiv 2023

  20. [21]

    Gregory Schraw and David Moshman. 1995. Metacognitive Theories. Educational Psychology Review 7 (12 1995), 351–371. https://doi.org/10.1007/BF02212307

    work page doi:10.1007/bf02212307 1995

  21. [22]

    Yueming Sun and Yi Zhang. 2018. Conversational recommender system. In The 41st international acm sigir conference on research & development in information retrieval . 235–244

    2018

  22. [23]

    Tuhina Tripathi, Manya Wadhwa, Greg Durrett, and Scott Niekum. 2025. Pairwise or Pointwise? Evaluating Feedback Protocols for Bias in LLM-Based Evaluation. arXiv preprint arXiv:2504.14716 (2025)

    work page arXiv 2025

  23. [24]

    Xiaolei Wang, Xinyu Tang, Wayne Xin Zhao, Jingyuan Wang, and Ji-Rong Wen. 2023. Rethinking the evaluation for conversational recommendation in the era of large language models. arXiv preprint arXiv:2305.13112 (2023). ACM Trans. Inf. Syst., Vol. 1, No. 1, Article 1. Publication date: January 2025. Towards Fast Domain Adaptation and Fine-Grained User Simula...

    work page arXiv 2023

  24. [25]

    Xiaolei Wang, Kun Zhou, Ji-Rong Wen, and Wayne Xin Zhao. 2022. Towards Unified Conversational Recommender Systems via Knowledge-Enhanced Prompt Learning. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’22) . ACM, 1929–1937. https://doi.org/10.1145/3534678.3539382

    work page doi:10.1145/3534678.3539382 2022

  25. [26]

    Hanwei Xu, Yujun Chen, Yulun Du, Nan Shao, Yanggang Wang, Haiyu Li, and Zhilin Yang. 2022. Gps: Genetic prompt search for efficient few-shot learning. arXiv preprint arXiv:2210.17041 (2022)

    work page arXiv 2022

  26. [27]

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. 2025. Qwen3 technical report. arXiv preprint arXiv:2505.09388 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  27. [28]

    Muchen Yang, Moxin Li, Yongle Li, Zijun Chen, Chongming Gao, Junqi Zhang, Yangyang Li, and Fuli Feng. 2024. Dual-Phase Accelerated Prompt Optimization. arXiv preprint arXiv:2406.13443 (2024)

    work page arXiv 2024

  28. [29]

    Lechen Zhang, Tolga Ergen, Lajanugen Logeswaran, Moontae Lee, and David Jurgens. 2024. Sprig: Improving large language model performance by system prompt optimization. arXiv preprint arXiv:2410.14826 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  29. [30]

    Shuo Zhang and Krisztian Balog. 2020. Evaluating conversational recommender systems via user simulation. In Proceedings of the 26th acm sigkdd international conference on knowledge discovery & data mining . 1512–1520

    2020

  30. [31]

    Xiaoyu Zhang, Ruobing Xie, Yougang Lyu, Xin Xin, Pengjie Ren, Mingfei Liang, Bo Zhang, Zhanhui Kang, Maarten de Rijke, and Zhaochun Ren. 2024. Towards empathetic conversational recommender systems. In Proceedings of the 18th ACM Conference on Recommender Systems . 84–93

    2024

  31. [32]

    Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric Xing, et al. 2023. Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in neural information processing systems 36 (2023), 46595–46623

    2023

  32. [33]

    Kun Zhou, Wayne Xin Zhao, Shuqing Bian, Yuanhang Zhou, Ji-Rong Wen, and Jingsong Yu. 2020. Improving conver- sational recommender systems via knowledge graph based semantic fusion. In Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining . 1006–1014

    2020

  33. [34]

    Yongchao Zhou, Andrei Ioan Muresanu, Ziwen Han, Keiran Paster, Silviu Pitis, Harris Chan, and Jimmy Ba. 2022. Large language models are human-level prompt engineers. In The eleventh international conference on learning repre- sentations

    2022

  34. [35]

    Lixi Zhu, Xiaowen Huang, and Jitao Sang. 2024. How reliable is your simulator? analysis on the limitations of current llm-based user simulators for conversational recommendation. In Companion Proceedings of the ACM Web Conference

    2024

  35. [36]

    Lixi Zhu, Xiaowen Huang, and Jitao Sang. 2024. A LLM-based Controllable, Scalable, Human-Involved User Simulator Framework for Conversational Recommender Systems. arXiv: 2405.08035 [cs.HC] https://arxiv.org/abs/2405.08035

    work page arXiv 2024

  36. [37]

    Lixi Zhu, Xiaowen Huang, and Jitao Sang. 2025. A llm-based controllable, scalable, human-involved user simulator framework for conversational recommender systems. In Proceedings of the ACM on Web Conference 2025 . 4653–4661. ACM Trans. Inf. Syst., Vol. 1, No. 1, Article 1. Publication date: January 2025

    2025