Pith. sign in

REVIEW 4 major objections 6 minor 58 references

Simulating Before Planning: Constructing Intrinsic User World Model for User-Tailored Dialogue Policy Planning

T0 review · 4 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read The paper claims that dialogue policy planning improves when the agent first builds an intrinsic model of the user—inferring the persona with a diffusion model and anticipating reactions with a Brownian Bridge—before choosing its next…

desk verdict Worth engaging with: a useful user-modeling framework and a new persona protocol, but the ESConv results conflict across tables and the active-learning rule looks backwards. read the letter →

arxiv 2504.13643 v1 pith:AOP7Q6MG submitted 2025-04-18 cs.CL

classification cs.CL
keywords User-tailoredDialoguePolicyPlanningLLM-basedAgentsusersimulationdiffusionmodelBrownianBridgepersonastrategypersuasionandemotionalsupportdialogues
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that dialogue policy planning has been optimizing the wrong variable: strategies are chosen against a generic user, while real users differ in personality, preferences, and decision style. To fix this, it introduces UDP, which first constructs an Intrinsic User World Model—a diffusion-based portrayer that infers the user's persona from the conversation so far, and a Brownian Bridge-based anticipator that predicts how that user would react to each candidate strategy—and then plans the next response using both. On self-play evaluations over 16 personas for donation persuasion (P4G) and 8 personas for emotional support (ESConv), UDP outperforms existing planners on success rate and soft success rate, with the largest gains on the hardest user types. The paper also shows that ignoring user traits leaves large persona-driven performance gaps, up to 19.2% on P4G and 34.9% on ESConv in success-rate variation across personas, which its framework narrows.

What carries the argument

The central object is the Intrinsic User World Model, made of two trained modules. The User Persona Portrayer is a conditional diffusion model: starting from pure noise at dialogue turn 0, it performs N/T denoising steps after each turn conditioned on the encoded user utterances, yielding a predicted persona distribution used both as a classification target and as the bridge endpoint for the next stage. The User Feedback Anticipator is a Brownian Bridge process conditioned on the inferred persona; its transition distribution predicts the distribution of the user's next latent state given the previous user reaction and the candidate system strategy, with variance that swells in the middle of the dialogue and shrinks near the fixed endpoints. Together they form a lightweight world model that lets the planner ask, before acting, what kind of user this is and how they will react if the agent does something.

What would settle it

Run the same P4G and ESConv evaluations with real human users playing the persona-defined roles instead of an LLM user simulator, keeping the same strategy labels and success criteria; if UDP's success-rate advantage over TRIP shrinks to noise, the intrinsic user world model is an artifact of LLM role-play rather than a faithful model of user traits. A cheaper first check is to have human annotators score, turn by turn, whether UDP's anticipated user reactions match what a real user of that persona later says.

Watch

Extended reading notes

Core claim

The central claim is that user-tailored policy planning succeeds by simulating before planning: a dialogue agent that can estimate who it is talking to and foresee the user's reaction to a strategy will pick better next actions than one that optimizes over a uniform user agent. UDP embodies this in three stages. Stage 1 treats persona inference as a conditional denoising task, walking a diffusion model from noise toward a clean persona representation as user utterances accumulate. Stage 2 models the possible next user state as a Brownian Bridge pinned between the current reaction and the terminal persona, so the uncertainty of the prediction is naturally highest mid-dialogue. Stage 3 feeds a fusion of dialogue history and inferred persona, together with strategy-plus-predicted-reaction features, into a softmax policy. The reported result is consistent and significant improvement over planners that ignore or weakly model user traits, for example success rate 0.598 versus 0.495 for TRIP on P4G and 0.832 versus 0.808 on ESConv, with the biggest wins on rejective, introvert, and neurotic personas.

Load-bearing premise

The load-bearing premise is that an LLM prompted with a persona description behaves like a real user with that persona, and that a critic LLM's scalar reward exceeding 0.6 measures true dialogue success.

Editorial extensions

If this is right

  • On both tasks, UDP's success-rate gains over TRIP are concentrated on the personas where baselines fail most—rejective, introvert, neurotic, and conservative—so modeling user traits is exactly what recovers performance on hard users.
  • The persona evaluation protocol exposes a 19.2% (P4G) and 34.9% (ESConv) spread in success rate across personas for existing planners, which means user diversity should be a standard axis in dialogue planning evaluation.
  • Ablations show that on the non-cooperative P4G task both the persona portrayer and the feedback anticipator contribute, while on the cooperative ESConv task the feedback anticipator can be removed without harm, so the value of reaction prediction depends on task cooperativity.
  • Active learning that up-weights personas after failed dialogues improves both metrics on P4G, indicating that the training distribution, not just the architecture, matters for user-tailored policies.
  • On ESConv, UDP deliberately comforts before suggesting and therefore uses more turns than TRIP; the paper interprets this as closer to real emotion-support practice rather than as an efficiency loss.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper's stated results, the same simulate-before-plan pipeline could transfer directly to conversational search and recommendation: pretrain the world model on persona-tagged interaction logs, then run candidate strategies through the feedback anticipator before the system commits to a response.
  • The Brownian Bridge's variance profile implies that user reactions are hardest to foresee mid-dialogue; a testable extension is to make the planner deliberately more exploratory or more clarifying in middle turns, and to check whether that improves success rate on hard personas.
  • The ESConv ablation, where removing the anticipator helps, suggests the world-model modules are not universally beneficial; a portfolio view—learning when to simulate reactions and when to rely on persona portraiture alone—is a natural next question.
  • A cheap validity check on the evaluation protocol: replace the critic's scalar reward with human success labels on a held-out set of simulated dialogues, and see whether UDP's margin over TRIP survives the swap.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper argues that existing dialogue policy planners rely on a uniform user simulation and therefore fail to adapt to diverse user traits. It first proposes a task-specific persona-based evaluation protocol for the P4G persuasion and ESConv emotional-support tasks, using GPT-generated profiles and LLM self-play with a critic reward to measure success. It then introduces UDP, a three-stage framework: a diffusion-based User Persona Portrayer that infers user traits from dialogue history, a Brownian Bridge-inspired User Feedback Anticipator that predicts user reactions to candidate strategies, and a User-Aware Policy Planner that combines history, persona, and anticipated feedback to select actions. UDP is trained by pretraining each stage on self-play data with persona labels, followed by policy-gradient reinforcement learning with an active-learning sampling scheme over personas. The paper reports consistent improvements over prior planners on both tasks, together with a human evaluation, per-persona analysis, ablation studies, and analyses of persona-prediction accuracy and strategy-distribution differences.

Significance. If the reported results hold, the paper makes a useful contribution by foregrounding user diversity in dialogue policy planning and by providing a concrete simulation protocol for measuring user-tailored behavior. The three-stage decomposition (persona portrayer, feedback anticipator, planner) is a sensible architectural idea, and the paper includes human evaluation, per-persona breakdowns, and several diagnostic analyses. However, the current validation is entirely internal to the authors' synthetic user world, and there are internal numerical inconsistencies, an asserted rather than derived stochastic-process formula, and a training objective that appears opposite to its stated purpose. These issues are load-bearing for the central claim of consistent, significant gains, so the manuscript needs substantive revision before the claims can be accepted.

major comments (4)
  1. [§5.2 Table 2 vs. §5.5 Table 4] The ESConv results for the full UDP model are inconsistent across tables: Table 2 reports SSR 0.774 / SR 0.832 (above TRIP's 0.744 / 0.808), while Table 4 reports SSR 0.742 / SR 0.791 for the same full model (below TRIP), and the '-w/o S2' row in Table 4 carries exactly the Table 2 full-model values (0.774 / 0.832). No note explains different checkpoints, evaluation splits, or random seeds. Because the headline claim of consistent ESConv improvements and the ablation conclusion that removing Stage 2 helps on ESConv both rest on these numbers, the discrepancy must be resolved before the main claim is reliable.
  2. [§4.3.4, Eq. (11)] The active-learning update is reversed relative to its stated goal. The text says the agent should 'trial more times with more challenging users' and 'prioritize challenging user personas,' but Eq. (11) increments w_i after a successful dialogue and decrements it after a failure. If w_i is a sampling weight, this increases sampling of easy personas and decreases sampling of hard ones. Please correct the update to increase weight on failures, or clarify that the intended objective is different and adjust the text accordingly. As written, the active-learning contribution and the '-w/o AL' ablation in §5.5 do not test the stated mechanism.
  3. [§4.2.2, Eq. (5)] Equation (5) is asserted rather than derived and deviates from the standard Brownian Bridge conditional distribution. For a bridge pinned at (t-1, z_{t-1}) and (T, z_T), the conditional distribution at time t has mean ((T-t)/(T-t+1)) z_{t-1} + (1/(T-t+1)) z_T and variance (T-t)/(T-t+1). The proposed mean inserts z_a into the first term, and the variance is 4(T-t)ψ(z_a)/(T-t+1)^2, with no derivation or justification of the factor 4 or the ψ scaling. The Brownian Bridge motivation, the pretraining objective in Eq. (8), and the downstream action features in §4.2.3 all depend on this distribution, so the formula needs to be derived or corrected.
  4. [§3.3, §4.3, §5.1.2] There is a partial circularity between training and evaluation: the same synthetic user world is used on both sides. The GPT-3.5-turbo-generated profiles condition the GPT-4o-mini user role-play, and the persona labels used to construct the user simulator are also the targets used to train the portrayer and the basis of the evaluation protocol. The human evaluation in §3.3.2 only verifies that role-play is consistent with the prompt, not that it resembles real users. To support the 'user-tailored' claim beyond simulation, please report results on held-out personas or on real interaction data, or explicitly restrict the claims to the simulated protocol and discuss how persona-conditioned simulation could bias the measured gains.
minor comments (6)
  1. [§5.2] The text says 'increasing the SRR and SR' and later 'improvements of SRR and SR'; the metric is SSR throughout the tables, so please correct the typo.
  2. [§4.1.2, Eq. (2)] The forward noising kernel is written as q(x_{i-1}|x_i) with x_i on the left and x_{i-1} on the right; the standard form is q(x_i|x_{i-1}) = N(x_i; sqrt(1-beta_i) x_{i-1}, beta_i I), so the conditioning and subscripts should be corrected.
  3. [§4.3.2, Eq. (8)] The denominator in the contrastive loss is typeset with an undefined symbol; please define the negative samples explicitly, for example as z' drawn uniformly from the batch, so that the loss is well defined.
  4. [References [12] and [13]] References [12] and [13] are the same paper (Fu et al., 2023), and Table 2 cites ICL-AIF as [13] while §5.1.1 cites it as [12]; please unify the citations.
  5. [§5.1.2 / Table 2 caption] The caption of Table 2 states that results are averaged over five inference runs with p < 0.05 under a t-test, but the table contains no significance markers, confidence intervals, or standard deviations; please add them or clarify which pairwise comparisons were tested.
  6. [§5.3, Table 3] The table header for the Identification dimension is 'Ind.' while the text abbreviates it as 'Ide.'; please make the abbreviation consistent.

Circularity Check

2 steps flagged · score 4.0 of 10

UDP's reported success partly reduces to its own RL objective and author-defined persona simulation, though held-out profiles and external baselines prevent full circularity.

  1. fitted input called prediction [Section 3.3.1 (self-play critic threshold) and Section 4.3.4, Eq. (10)]
    "Following previous work [19], we view the dialogue objective as being achieved if r(s_t) exceeds a threshold of 0.6. ... L_rl(θ) = − Σ_t log π_θ(a_t|s_{t−1}) · R_t, where R_t = Σ_{t′=t}^{T} γ^{T−t′} r_{t′} represents the cumulative reward."

    The headline SR and SSR metrics are computed from the same GPT-3.5-turbo critic reward r_t that Eq. (10) directly maximizes during reinforcement learning: SR is the fraction of episodes whose final critic reward exceeds 0.6, and SSR is the mean final critic reward. The policy is therefore fitted to the very scalar that is later reported as 'success rate,' so the reported success is partly an in-sample check of the optimized objective rather than an independent measurement. The comparison with baselines may still be fair, but the central external-validity claim is not derived from evidence independent of the training signal.

  2. self definitional [Section 3.1–3.2 and Section 4.3, with Section 4.2.1]
    "These dialogues were prompted using the predefined user profiles; thus, the constructed pretraining dataset inherently contains user persona labels. ... During evaluation, we select the user persona with the highest probability from D_t to help subsequent dialogue policy planning."

    Stage 1 is supervised with the persona label y_p that the authors attached to each generated profile, and the same persona taxonomy is used to prompt the GPT-4o-mini user during both self-play training and evaluation. The model is therefore trained to recover the authors' own construction from behavior generated by that construction, and the user-tailored planning advantage is demonstrated only against users defined by the same labels. The personas are not grounded in an independent external population, so the 'intrinsic user world model' is a function fitted to the authors' simulation rather than an independently validated model of real users.

full rationale

The central derivation is not fully circular because UDP is compared against several external baselines on held-out profiles, and the persona labels are hidden from the model during inference, requiring genuine inference from dialogue behavior. However, two structural loops weaken the independence of the main claim. First, the RL objective in Eq. (10) maximizes the same critic reward that defines SR/SSR in Section 3.3.1, so the reported success metric is partly the optimized training signal. Second, the persona taxonomy and profile generator are author-defined inputs that simultaneously create the training labels, the test user behavior, and the 'world model' targets; the evaluation therefore measures performance inside the authors' synthetic construction. The internal inconsistency between Table 2 and Table 4 on ESConv is primarily a correctness and reproducibility concern rather than a circularity finding, but it further undermines the headline claim of consistent gains. Overall, the paper has partial circularity rather than definitional identity, so a score of 4 is appropriate.

Assumptions & free parameters 3 free parameters · 5 assumptions · 2 invented entities

The whole evaluation is a closed synthetic loop: the authors design personas, generate profiles with GPT-3.5-turbo, have GPT-4o-mini role-play them, and score success with GPT-3.5-turbo. Training and testing happen inside this loop, so the only external anchor is the held-out persona split. The headline metrics depend on a hand-set reward mapping and a 0.6 threshold, and the Brownian Bridge feedback module rests on an ad hoc trajectory assumption. There are no real-user interactions and no external behavioral benchmarks, so the generalizability claims are not independently grounded.

free parameters (3)
  • Critic reward mapping and success threshold = P4G: -1.0/-0.5/0.1/1.0; ESConv: -1.0/-0.5/0.1/1.0/1.0; threshold 0.6
    Hand-set mapping and threshold determine all SR/SSR values; no sensitivity analysis is provided (Section 3.3.1).
  • Brownian Bridge variance scale = 4 (Eq.5)
    The factor 4 in the variance of Eq.5 is arbitrary and not derived from the Brownian Bridge equation (Section 4.2.2).
  • Maximum dialogue turns T = 10
    Endpoint of the Brownian Bridge and termination condition for self-play; standard in prior work but hand-selected (Section 5.1.2).
assumptions (5)
  • standard math Brownian Bridge conditional distribution property (Revuz and Yor 1990).
    Used in Section 4.1.3 to motivate modeling user feedback as a bridge; Eq.5 then modifies the formula, so the standard result is not used as stated.
  • domain assumption A small set of hand-crafted binary traits captures the relevant diversity of real users in P4G and ESConv.
    Personas are combinations of three binary traits (16 for P4G, 8 for ESConv) designed by the authors in Section 3.1; no evidence that these axes cover real user diversity.
  • domain assumption LLM role-play with persona profiles faithfully simulates real users with those traits.
    All training and test dialogues come from GPT-4o-mini self-play with author-written prompts; human evaluation in Section 3.3.2 checks consistency with the prompt, not with real-user behavior.
  • domain assumption The LLM critic's reward mapping and 0.6 threshold correctly measure dialogue success.
    The critic is GPT-3.5-turbo with hand-set option values; no validation of these rewards against real task outcomes is given (Section 3.3.1).
  • ad hoc to paper User feedback states follow a modified Brownian Bridge depending on the action and a learned variance function.
    Eq.5 is presented as derived from Eq.4 but introduces action-dependent perturbation and learned MLP variance; this trajectory prior is specific to this paper (Section 4.2.2).
invented entities (2)
  • Latent user feedback state z_t
    purpose: Represents the predicted next user reaction in the feedback anticipator and conditions the policy planner.
    Learned latent with no external validation; only compared to encoded LLM user utterances during pretraining.
  • Persona embeddings P=[p_i]
    purpose: Serve as Brownian Bridge endpoints and classifier targets for user persona inference.
    Created by encoding author-written persona prompts with a frozen RoBERTa model; no real-user dataset grounds them.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Simulating Before Planning: Constructing Intrinsic User World Model for User-Tailored Dialogue Policy Planning." pith.science (2026). https://pith.science/paper/AOP7Q6MG

@misc{pith2026250413643,
  author       = {Pith},
  title        = {Pith review of: Simulating Before Planning: Constructing Intrinsic User World Model for User-Tailored Dialogue Policy Planning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AOP7Q6MG}},
  note         = {Machine review of arXiv:2504.13643}
}
read the original abstract

Recent advancements in dialogue policy planning have emphasized optimizing system agent policies to achieve predefined goals, focusing on strategy design, trajectory acquisition, and efficient training paradigms. However, these approaches often overlook the critical role of user characteristics, which are essential in real-world scenarios like conversational search and recommendation, where interactions must adapt to individual user traits such as personality, preferences, and goals. To address this gap, we first conduct a comprehensive study utilizing task-specific user personas to systematically assess dialogue policy planning under diverse user behaviors. By leveraging realistic user profiles for different tasks, our study reveals significant limitations in existing approaches, highlighting the need for user-tailored dialogue policy planning. Building on this foundation, we present the User-Tailored Dialogue Policy Planning (UDP) framework, which incorporates an Intrinsic User World Model to model user traits and feedback. UDP operates in three stages: (1) User Persona Portraying, using a diffusion model to dynamically infer user profiles; (2) User Feedback Anticipating, leveraging a Brownian Bridge-inspired anticipator to predict user reactions; and (3) User-Tailored Policy Planning, integrating these insights to optimize response strategies. To ensure robust performance, we further propose an active learning approach that prioritizes challenging user personas during training. Comprehensive experiments on benchmarks, including collaborative and non-collaborative settings, demonstrate the effectiveness of UDP in learning user-specific dialogue strategies. Results validate the protocol's utility and highlight UDP's robustness, adaptability, and potential to advance user-centric dialogue systems.

Figures

Figures reproduced from arXiv: 2504.13643 by the authors.

Figure 1
Figure 1. Framework overview. The framework UDP consists [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The portraying (denoising) process of the Diffusion [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. The architecture comparison of the Policy Planning [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Firstly, UDP outperforms baselines across most user types [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Changes in user persona prediction accuracy with [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

58 extracted references · 26 canonical work pages

  1. [1]

    Layla El Asri, Jing He, and Kaheer Suleman. 2016. A Sequence-to-Sequence Model for User Simulation in Spoken Dialogue Systems. In Interspeech. https: //arxiv.org/pdf/1607.00070

  2. [2]

    Do, Yan Xu, and Pascale Fung

    Yejin Bang, Samuel Cahyawijaya, Nayeon Lee, Wenliang Dai, Dan Su, Bryan Wilie, Holy Lovenia, Ziwei Ji, Tiezheng Yu, Willy Chung, Quyet V. Do, Yan Xu, and Pascale Fung. 2023. A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity. ArXiv abs/2302.04023 (2023). https://aclanthology.org/2023.ijcnlp-main.45.pdf

  3. [3]

    Jiangjie Chen, Xintao Wang, Rui Xu, Siyu Yuan, Yikai Zhang, Wei Shi, Jian Xie, Shuang Li, Ruihan Yang, Tinghui Zhu, Aili Chen, Nianqi Li, Lida Chen, Caiyu Hu, Siye Wu, Scott Ren, Ziquan Fu, and Yanghua Xiao. 2024. From Persona to Per- sonalization: A Survey on Role-Playing Language Agents. ArXiv abs/2404.18231 (2024). https://arxiv.org/pdf/2404.18231

  4. [4]

    Nuo Chen, Yang Deng, and Jia Li. 2024. The Oscars of AI Theater: A Survey on Role-Playing with Language Models. ArXiv abs/2407.11484 (2024). https: //arxiv.org/pdf/2407.11484

  5. [5]

    Hsien chin Lin, Shutong Feng, Christian Geishauser, Nurul Lubis, Carel van Niekerk, Michael Heck, Benjamin Matthias Ruppik, Renato Vukovic, and Milica Gavsi’c. 2023. EmoUS: Simulating User Emotions in Task-Oriented Dialogues. Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval (2023). https://dl....

  6. [6]

    Konstantina Christakopoulou, Alex Beutel, Rui Li, Sagar Jain, and Ed H Chi. 2018. Q&R: A two-stage approach toward interactive recommendation. In Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 139–148

  7. [7]

    Le, and Lizi Liao

    Huy Dao, Yang Deng, Khanh-Huyen Bui, Dung D. Le, and Lizi Liao. 2024. Ex- perience as Source for Anticipation and Planning: Experiential Policy Learning for Target-driven Recommendation Dialogues. In Conference on Empirical Meth- ods in Natural Language Processing . https://api.semanticscholar.org/CorpusID: 273822277

  8. [8]

    Yang Deng, Wenqiang Lei, Hongru Wang, and Tat seng Chua. 2023. Prompting and Evaluating Large Language Models for Proactive Dialogues: Clarification, Target-guided, and Non-collaboration. In Conference on Empirical Methods in Natural Language Processing. https://arxiv.org/pdf/2305.13626

Show all 58 references
  1. [9]

    Yang Deng, Wenxuan Zhang, Wai Lam, See-Kiong Ng, and Tat-Seng Chua. 2023. Plug-and-Play Policy Planner for Large Language Model Powered Dialogue Agents. ArXiv abs/2311.00262 (2023). https://arxiv.org/pdf/2311.00262

  2. [10]

    Prafulla Dhariwal and Alex Nichol. 2021. Diffusion Models Beat GANs on Image Synthesis. ArXiv abs/2105.05233 (2021). https://arxiv.org/pdf/2105.05233

  3. [11]

    Jingtao Ding, Yunke Zhang, Yu Shang, Yuheng Zhang, Zefang Zong, Jie Feng, Yuan Yuan, Hongyuan Su, Nian Li, Nicholas Sukiennik, Fengli Xu, and Yong Li

  4. [13]

    Yao Fu, Hao-Chun Peng, Tushar Khot, and Mirella Lapata. 2023. Improving Language Model Negotiation with Self-Play and In-Context Learning from AI Feedback. ArXiv abs/2305.10142 (2023). https://arxiv.org/pdf/2305.10142

  5. [14]

    Jianfeng Gao, Michel Galley, and Lihong Li. 2018. Neural Approaches to Conversa- tional AI. The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval (2018). https://dl.acm.org/doi/10.1145/3209978.3210183

  6. [15]

    Goldberg

    Lewis R. Goldberg. 1992. THE DEVELOPMENT OF MARKERS FOR THE BIG- FIVE FACTOR STRUCTURE. Psychological Assessment 4 (1992), 26–42. https: //psycnet.apa.org/doiLanding?doi=10.1037/1040-3590.4.1.26

  7. [16]

    Shasha Guo, Lizi Liao, Jing Zhang, Cuiping Li, and Hong Chen. 2024. PCQPR: Proactive Conversational Question Planning with Reflection. In Conference on Empirical Methods in Natural Language Processing. https://aclanthology.org/2024. emnlp-main.631.pdf

  8. [17]

    David R Ha and Jürgen Schmidhuber. 2018. World Models. ArXiv abs/1803.10122 (2018). https://arxiv.org/pdf/1803.10122

  9. [18]

    Tao He, Lizi Liao, Yixin Cao, Yuanxing Liu, Ming Liu, Zerui Chen, and Bing Qin

  10. [19]

    Tao He, Lizi Liao, Yixin Cao, Yuanxing Liu, Yiheng Sun, Zerui Chen, Ming Liu, and Bing Qin. 2024. Simulation-Free Hierarchical Latent Policy Planning for Proactive Dialogues. https://arxiv.org/pdf/2412.14584

  11. [20]

    ArXiv abs/2406.05374 (2024)

    Planning Like Human: A Dual-process Framework for Dialogue Planning. ArXiv abs/2406.05374 (2024). https://arxiv.org/pdf/2406.05374

  12. [21]

    Campbell, K

    Lukasz Kaiser, Mohammad Babaeizadeh, Piotr Milos, Blazej Osinski, Roy H. Campbell, K. Czechowski, D. Erhan, Chelsea Finn, Piotr Kozakowski, Sergey Levine, Afroz Mohiuddin, Ryan Sepassi, G. Tucker, and Henryk Michalewski

  13. [22]

    Jonathan Ho, Ajay Jain, and P. Abbeel. 2020. Denoising Diffusion Probabilistic Models. ArXiv abs/2006.11239 (2020). https://arxiv.org/pdf/2006.11239

  14. [23]

    Yann LeCun and Courant. 2022. A Path Towards Autonomous Machine Intel- ligence Version 0.9.2, 2022-06-27. https://api.semanticscholar.org/CorpusID: 251881108

  15. [24]

    Li, Hasan Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem

    G. Li, Hasan Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem

  16. [25]

    Wai-Chung Kwan, Hongru Wang, Huimin Wang, and Kam-Fai Wong. 2022. A Survey on Recent Advances and Challenges in Reinforcement Learning Methods for Task-oriented Dialogue Policy Learning. Machine Intelligence Research (2022), 1–17. https://link.springer.com/article/10.1007/s116...

  17. [26]

    Jinggui Liang, Lizi Liao, Hao Fei, Bobo Li, and Jing Jiang. 2024. Actively Learn from LLMs with Uncertainty Propagation for Generalized Category Discovery. In North American Chapter of the Association for Computational Linguistics . https: //aclanthology.org/2024.naacl-long.434.pdf

  18. [27]

    Hsien-Chin Lin, Christian Geishauser, Shutong Feng, Nurul Lubis, Carel van Niekerk, Michael Heck, and Milica Gavsi’c. 2022. GenTUS: Simulating User Be- haviour and Language in Task-oriented Dialogues with Generative Transformers. ArXiv abs/2208.10817 (2022). https://aclantholo...

  19. [28]

    Hsien-Chin Lin, Nurul Lubis, Songbo Hu, Carel van Niekerk, Christian Geishauser, Michael Heck, Shutong Feng, and Milica Gavsi’c. 2021. Domain-independent User Simulation with Transformers for Task-oriented Dialogue Systems. In SIGDIAL Conferences. https://aclanthology.org/2021...

  20. [29]

    Williams

    Lihong Li, He He, and J. Williams. 2014. Temporal supervised learning for inferring a dialog policy from example conversations. 2014 IEEE Spoken Lan- guage Technology Workshop (SLT) (2014), 312–317. https://ieeexplore.ieee.org/ document/7078593/

  21. [30]

    Keming Lu, Bowen Yu, Chang Zhou, and Jingren Zhou. 2024. Large Language Models are Superpositions of All Characters: Attaining Arbitrary Role-play via Self-Alignment. ArXiv abs/2401.12474 (2024). https://arxiv.org/pdf/2401.12474

  22. [31]

    María Luisa Menéndez, JA Pardo, L Pardo, and MC Pardo. 1997. The jensen- shannon divergence. Journal of the Franklin Institute 334, 2 (1997), 307–318

  23. [32]

    Filip Moons and Ellen Vandervieren. 2023. Measuring agreement among several raters classifying subjects into one-or-more (hierarchical) nominal categories. A generalisation of Fleiss’ kappa. https://arxiv.org/pdf/2303.12502

  24. [33]

    Siyang Liu, Chujie Zheng, Orianna Demasi, Sahand Sabour, Yu Li, Zhou Yu, Yong Jiang, and Minlie Huang. 2021. Towards Emotional Support Dialog Systems. In Annual Meeting of the Association for Computational Linguistics . https:// aclanthology.org/2021.acl-long.269.pdf

  25. [34]

    Daniel Revuz and Marc Yor. 1990. Continuous martingales and Brownian motion. https://link.springer.com/book/10.1007/978-3-662-21726-9

  26. [35]

    Scott and Reginald A

    Susanne G. Scott and Reginald A. Bruce. 1995. Decision-Making Style: The Development and Assessment of a New Measure. Educational and Psychological Measurement 55 (1995), 818 – 831. https://api.semanticscholar.org/CorpusID: 143479230

  27. [36]

    Ivan Sekuli’c, Silvia Terragni, Victor Guimaraes, Nghia Khau, Bruna Guedes, Modestas Filipavicius, Andre Manso, and Roland Mathis. 2024. Reliable LLM- based User Simulator for Task-Oriented Dialogue Systems. ArXiv abs/2402.13374 (2024). https://aclanthology.org/2024.scichat-1.3.pdf

  28. [37]

    Baolin Peng, Xiujun Li, Jianfeng Gao, Jingjing Liu, and Kam-Fai Wong. 2018. Deep Dyna-Q: Integrating Planning for Task-Completion Dialogue Policy Learning. In Annual Meeting of the Association for Computational Linguistics . https:// aclanthology.org/P18-1203.pdf

  29. [38]

    Sutton, David A

    Richard S. Sutton, David A. McAllester, Satinder Singh, and Y. Mansour. 1999. Policy Gradient Methods for Reinforcement Learning with Function Approxi- mation. In Neural Information Processing Systems . https://www.cis.upenn.edu/ ~mkearns/finread/Sutton.pdf

  30. [39]

    Leila Tavakoli, Hamed Zamani, Falk Scholer, William Bruce Croft, and Mark Sanderson. 2022. Analyzing clarification in asynchronous information-seeking conversations. Journal of the Association for Information Science and Technology 73, 3 (2022), 449–471

  31. [40]

    Quan Tu, Shilong Fan, Zihang Tian, and Rui Yan. 2024. CharacterEval: A Chinese Benchmark for Role-Playing Conversational Agent Evaluation. InAnnual Meeting of the Association for Computational Linguistics . https://arxiv.org/pdf/2401.01275

  32. [41]

    Jiaming Song, Chenlin Meng, and Stefano Ermon. 2020. Denoising Diffusion Implicit Models. ArXiv abs/2010.02502 (2020). https://arxiv.org/pdf/2010.02502

  33. [42]

    Jian Wang, Dongding Lin, and Wenjie Li. 2023. Dialogue Planning via Brow- nian Bridge Stochastic Process for Goal-directed Proactive Dialogue. ArXiv abs/2305.05290 (2023). https://arxiv.org/pdf/2305.05290

  34. [43]

    Xuewei Wang, Weiyan Shi, Richard Kim, Yoo Jung Oh, Sijia Yang, Jingwen Zhang, and Zhou Yu. 2019. Persuasion for Good: Towards a Personalized Persuasive Dialogue System for Social Good. ArXiv abs/1906.06725 (2019). https://aclanthology.org/P19-1566.pdf

  35. [44]

    Zekun Moore Wang, Zhongyuan Peng, Haoran Que, Jiaheng Liu, Wangchun- shu Zhou, Yuhan Wu, Hongcheng Guo, Ruitong Gan, Zehao Ni, Man Zhang, Simulating Before Planning: Constructing Intrinsic User World Model for User-Tailored Dialogue Policy Planning SIGIR ’25, July 13-18, 2025,...

  36. [45]

    Stefan Ultes, Lina Maria Rojas-Barahona, Pei hao Su, David Vandyke, Dongho Kim, Iñigo Casanueva, Paweł Budzianowski, Nikola Mrksic, Tsung-Hsien Wen, Milica Gavsic, and Steve J. Young. 2017. PyDial: A Multi-domain Statistical Dialogue System Toolkit. In Annual Meeting of the As...

  37. [46]

    Jianxiang Xiang, Zhenhua Liu, Haodong Liu, Yin Bai, Jia Cheng, and Wenliang Chen. 2024. DiffusionDialog: A Diffusion Model for Diverse Dialog Genera- tion with Latent Space. In International Conference on Language Resources and Evaluation. https://aclanthology.org/2024.lrec-ma...

  38. [47]

    Jing Ye, Lu Xiang, Yaping Zhang, and Chengqing Zong. 2024. SweetieChat: A Strategy-Enhanced Role-playing Framework for Diverse Scenarios Handling Emotional Support Agent. ArXiv abs/2412.08389 (2024). https://arxiv.org/pdf/ 2412.08389

  39. [48]

    Fengji Yi, Wenlong Fu, and Huan Liang. 2018. Model-based reinforcement learn- ing: A survey. https://api.semanticscholar.org/CorpusID:108339287

  40. [49]

    In Annual Meeting of the Association for Computational Linguistics

    RoleLLM: Benchmarking, Eliciting, and Enhancing Role-Playing Abilities of Large Language Models. In Annual Meeting of the Association for Computational Linguistics. https://arxiv.org/pdf/2310.00746

  41. [50]

    Bowen Wu, Kaili Sun, Ziwei Bai, Ying Li, and Baoxun Wang. 2025. RAIDEN Benchmark: Evaluating Role-playing Conversational Agents with Measurement- Driven Custom Dialogues. In Proceedings of the 31st International Conference on Computational Linguistics. 11086–11106

  42. [51]

    Shuo Zhang and Krisztian Balog. 2020. Evaluating Conversational Recommender Systems via User Simulation. Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (2020). https://dl.acm.org/ doi/10.1145/3394486.3403202

  43. [52]

    Tong Zhang, Chen Huang, Yang Deng, Hongru Liang, Jia Liu, Zujie Wen, Wen- qiang Lei, and Tat-Seng Chua. 2024. Strength Lies in Differences! Improving Strategy Planning for Non-collaborative Dialogues via Diversified User Sim- ulation. In Conference on Empirical Methods in Natu...

  44. [53]

    Zhirui Zhang, Xiujun Li, Jianfeng Gao, and Enhong Chen. 2019. Budgeted Policy Learning for Task-Oriented Dialogue Systems. In Annual Meeting of the Association for Computational Linguistics . https://aclanthology.org/P19-1364.pdf

  45. [54]

    Xiao Yu, Maximillian Chen, and Zhou Yu. 2023. Prompt-Based Monte-Carlo Tree Search for Goal-Oriented Dialogue Policy Planning. In Conference on Empirical Methods in Natural Language Processing . https://arxiv.org/pdf/2305.13660

  46. [55]

    Qiang Zhang, Jason Naradowsky, and Yusuke Miyao. 2023. Ask an Expert: Lever- aging Language Models to Improve Strategic Reasoning in Goal-Oriented Dia- logue Models. In Annual Meeting of the Association for Computational Linguistics . https://arxiv.org/pdf/2305.17878

  47. [59]

    Weixiang Zhao, Yanyan Zhao, Xin Lu, Shilong Wang, Yanpeng Tong, and Bing Qin. 2023. Is ChatGPT Equipped with Emotional Dialogue Capabilities? ArXiv abs/2304.09582 (2023). https://arxiv.org/pdf/2304.09582

  48. [2019]

    ArXiv abs/1903.00374 (2019)

    Model-Based Reinforcement Learning for Atari. ArXiv abs/1903.00374 (2019). https://arxiv.org/pdf/1903.00374

  49. [2023]

    In Neural Information Processing Systems

    CAMEL: Communicative Agents for "Mind" Exploration of Large Language Model Society. In Neural Information Processing Systems . https://arxiv.org/pdf/ 2303.17760

  50. [2024]

    https://arxiv.org/pdf/2411.14499

    Understanding World or Predicting Future? A Comprehensive Survey of World Models. https://arxiv.org/pdf/2411.14499

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.