Pith. sign in

REVIEW 5 major objections 4 minor 1 cited by

DiaCBT: A Long-Periodic Dialogue Corpus Guided by Cognitive Conceptualization Diagram for CBT-based Psychological Counseling

T0 review · 5 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a multi-session dialogue corpus guided by cognitive conceptualization diagrams lets a fine-tuned LLM outperform six counseling baselines on success rate, session length, positive-affect change, and CBT skill scores.

desk verdict A genuinely new multi-session CBT dialogue corpus with careful construction, but the headline claim leans almost entirely on an evaluation loop whose clients and judges share the same CCD framework as the training data. read the letter →

arxiv 2509.02999 v1 pith:6PAMUECY submitted 2025-09-03 cs.CL

Yougen Zhou , Ningning Zhou , Qin Chen , Jie Zhou , Aimin Zhou , Liang He This is my paper
classification cs.CL
keywords cognitivebehavioraltherapydialoguecorpusconceptualizationdiagramclientsimulationLLMfine-tuningmulti-sessioncounselingpsychotherapydatasetstrategyannotation
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 claims that the main obstacle to CBT-capable conversational agents is not model size but training data: existing counseling datasets compress an entire therapy into one session and give clients only shallow background descriptions. DiaCBT instead reconstructs full five-session CBT courses, annotates each therapist turn with one of fourteen CBT strategies, and models each client with a cognitive conceptualization diagram linking core beliefs, intermediate beliefs, situations, automatic thoughts, emotions, and behaviors. A Qwen2.5-7B model fine-tuned on this corpus achieves the highest success rate (77.14%), longest average session (12.05 turns), largest positive-affect gain (+1.675), and highest total Cognitive Therapy Rating Scale score (28.75) among all baselines, and human judges prefer its dialogues over two strong CBT baselines. The paper concludes that CCD-guided, multi-session, strategy-annotated dialogues are what let LLMs emulate CBT psychologists rather than give generic empathy or advice.

What carries the argument

The Cognitive Conceptualization Diagram (CCD), a six-component structured profile of a client's cognitive model (core beliefs, intermediate beliefs, situation, automatic thoughts, emotions, behaviors), is the load-bearing object. It simultaneously drives client simulation during evaluation—each simulated client is initialized from a CCD and instructed to reveal deeper concerns gradually—and guides dialogue generation during data construction, so that client utterances stay cognitively consistent across five sessions. The second mechanism is the 14-strategy CBT annotation scheme, which lets the fine-tuned model be trained to predict both the next strategy and the therapist utterance, and whic

What would settle it

Run the same fine-tuned model against human standardized patients—actors trained to present the same CCD profiles and to disclose gradually—and compare success rate, session length, and PANAS change with the GPT-4o-simulated-client scores; if the human-interaction scores fall well short of the simulated ones, the reported gains reflect simulation cooperativeness rather than counseling skill. A cheaper check: instruct the simulated client prompt to refuse disclosure and test whether success rate and CTRS collapse.

Watch

Extended reading notes

Core claim

DiaCBT is a long-periodic CBT counseling corpus of 108 cases, 540 sessions, and roughly 265 utterances per case, built by taking 53 real CBT transcripts, using LLMs to expand them into multiple sessions guided by a six-component Cognitive Conceptualization Diagram per client, and filtering the output with trained annotators and a licensed CBT therapist. The central claim is that a model fine-tuned on this corpus acquires CBT-specific competence that generic empathetic or single-session counseling models lack: it sustains longer sessions, succeeds more often in completing the counseling goal, raises client positive affect more, and scores higher on CBT-specific skill dimensions such as Guided

Load-bearing premise

GPT-4o-simulated clients, guided only by a cognitive conceptualization diagram, behave enough like real clients that success in simulation predicts success in real counseling; if simulated clients are too cooperative or the CCD omits real cognitive dynamics, the reported success rates and emotional-change scores overstate the model's clinical capability.

Editorial extensions

If this is right

  • A fine-tuned 7B model can sustain five-session CBT structure with correct strategy progression, suggesting multi-session datasets are sufficient to teach long-horizon counseling behavior without larger backbones.
  • Strategy annotations carry real signal: the annotated model asks far more in-depth questions (7.41 per session versus 2.75 for the unfine-tuned model), shifting output from generic advice toward reflective, client-led discovery.
  • The per-session strategy distribution can serve as a normative blueprint—for automated agenda-setting, session staging, or counselor-training curricula—since it mirrors the documented CBT sequence.
  • CCD-guided client simulation offers a reusable evaluation protocol for therapy agents that does not require recruiting at-risk human clients, and the reported expert ratings indicate the simulated clients display coherent, distinguishable cognitive profiles.
  • Because the corpus is built entirely from public cases and crowdsourced thought data, it provides a privacy-safe training resource for a domain where real clinical transcripts are largely inaccessible.

Reading between the lines

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

  • If CCD-guided simulation is as faithful as the expert ratings suggest, the same construction pipeline—structured case-formulation diagram plus strategy-annotated multi-session scripts—should transfer to other structured therapies (e.g., DBT or schema therapy) by swapping the diagram for that modality's case-formulation template.
  • The evaluation protocol could be stress-tested for robustness by varying the simulated client's cooperativeness, disclosure depth, or initial cognitive rigidity; a model that still succeeds with deliberately resistant clients would be stronger evidence of genuine skill rather than simulation compliance.
  • The paper leaves untested whether the learned behavior generalizes to presenting problems outside the seven C2D2 cognitive-distortion categories; a held-out set of novel situations and beliefs would reveal whether the model learned CBT technique or memorized dialogue patterns.
  • Since the model's negative-affect reduction (+1.021 drop) lags its positive-affect gains (+1.675), a plausible next target is designing strategies—or data—specifically aimed at de-escalation, a dimension the current corpus appears to under-weight.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 4 minor

Summary. The paper introduces DiaCBT, a CBT-based counseling dialogue corpus in Chinese, comprising 108 cases and 540 sessions, with each case structured across five sessions and annotated with 14 CBT strategies. A distinctive feature is the use of Cognitive Conceptualization Diagrams (CCDs) to guide client simulation during data generation. The authors fine-tune Qwen2.5-7B-Instruct on DiaCBT and evaluate it against several counseling baselines using simulated client interactions: GPT-4o acts as the client, another LLM provides reward-based success judgments, and LLM-based scoring is used for PANAS and CTRS. The paper reports that the fine-tuned model achieves the best success rate, average turns, positive affect change, and total CTRS, and it also presents a human preference evaluation. The central claim is that DiaCBT effectively enhances LLMs' ability to emulate psychologists with CBT expertise.

Significance. If the evaluation were externally valid, DiaCBT would be a valuable resource for CBT dialogue research: it is multi-session, strategy-annotated, and CCD-guided, and the data construction pipeline includes expert screening and refinement, with a reported retention rate of 72.58%. The authors also provide a human evaluation with substantial inter-rater agreement (Fleiss' kappa = 0.685). However, the load-bearing evaluation is almost entirely self-referential: the training dialogues, the simulated clients, the success judge, and the clinical rating scales all instantiate the same CCD-based cognitive framework. The paper does not demonstrate that success in this closed simulation loop transfers to real counseling ability. Because the central claim rests on this unvalidated loop, the significance of the empirical results is conditional on additional external validation.

major comments (5)
  1. [§5.1, §C.1, §C.2.1] The evaluation loop is partially circular. Training dialogues in §3.2–3.3 are generated from CCDs, and evaluation clients in §5.1/C.1 are initialized from newly constructed CCDs with a prompt that instructs GPT-4o to 'allow your responses to be informed by the diagram' and 'gradually reveal deeper concerns.' The success metric is determined by LLMrwd, a grader LLM given a goal-oriented rubric, and PANAS/CTRS are LLM-scored. A model that learns CCD-consistent question patterns can therefore achieve high scores even without producing genuine therapeutic change. The human evaluation in §5.4 uses the same kind of simulated clients, so it does not break this loop. The stated limitations and ethical considerations do not address this validity threat. To support the claimed enhancement of 'CBT expertise,' the authors should either (a) evaluate against human role-players not given the CCD, or re
  2. [§C.2.1] The success rate metric is not reproducible as defined. The text states that GOAL-COMPLETED is declared when the reward value v_t is 'not less than a certain threshold ε,' but the value of ε is never given. Likewise, 'predefined maximum number of turns' is mentioned without a concrete number. These are free parameters of the central evaluation metric, and without them, the reported SR values in Table 4 cannot be independently replicated or compared across settings.
  3. [Tables 4–5] The quantitative comparisons are reported as single point estimates with no variance, confidence intervals, or significance tests. For example, the total CTRS difference between Ours (28.75) and PsyChat (27.75) is one point on a scale where subscales are scored 0–6; without repeated runs, bootstrap intervals, or inter-judge variability on the LLM-scored metrics, these differences may reflect noise. Differences in SR (e.g., 77.14% vs. 72.85%) are similarly unaccompanied by uncertainty. The authors should provide variance estimates, e.g., across multiple evaluation seeds or bootstrap resamples of the 140 cases.
  4. [§5.4, Table 6] The human evaluation is a useful sanity check but does not validate the clinical-transfer claim because the annotators compare dialogues produced against the same GPT-4o simulated clients. Furthermore, the summary overstates the result: on the 'Relevance' criterion, CAMEL wins 89 vs. 51, which is a substantial loss, not a 'nearly all aspects' success. The authors should acknowledge this asymmetry and report tie counts and per-criterion agreement more transparently.
  5. [§3.1 vs. Table 1] There is an apparent provenance inconsistency. Section 3.1 says 53 transcripts were collected and divided into 33 sessions in 6 complete consultation processes, while Table 1 reports 108 cases and 540 sessions. The paper does not explain how 108 cases were derived from the 53 collected transcripts, nor what role the C2D2 and PatternReframe resources play in creating the final case count. If the 108 cases are synthetic expansions of the 6 complete processes, the claim that the data is 'well aligned with real-world scenarios' needs to be qualified and the scaling procedure documented.
minor comments (4)
  1. [Figure 7 prompt] The prompt contains a typo: 'metal health challenges' should be 'mental health challenges.'
  2. [Table 4 and §D.1] The backbone name is written inconsistently: 'InterLM2-Chat-7B' in Table 4 but 'InternLM2-7B-Chat' in Section D.1. Please unify the spelling.
  3. [Figure 4 caption] The figure caption appears corrupted in the manuscript text, with embedded '/uni0000...' sequences. The caption should be cleaned so that the rating distribution is interpretable.
  4. [§5.4] The Fleiss' kappa value 0.685 is reported, but the number of dialogues per annotator and the criteria-specific agreement are not fully reported. Providing a breakdown would strengthen the human evaluation.

Circularity Check

1 steps flagged · score 2.0 of 10

One self-consistency check for the client simulator is circular, but the central dataset/training comparison is not.

  1. self definitional [Section 5.5 (Analysis of LLMs as Client Simulators) and Figure 7 (Appendix C.1)]
    "Your thought process should be guided by the cognitive conceptualization diagram in the 'Cognitive Conceptualization Diagram' section, but avoid directly referencing the diagram as a real patient would not explicitly think in those terms. ... Instead, allow your responses to be informed by the diagram, enabling the therapist to infer you thought processes. [Figure 7] ... Three experts evaluate the simulated clients’ overall accuracy and the accuracy of each component in CCD. ... These findings validate the reliability of leveraging LLMs as client simulators, demonstrating their capability to a"

    The simulated client is generated by prompting GPT-4o with the CCD and explicitly instructing it to let responses be informed by that diagram. The validation in §5.5 then rates whether the simulated client's utterances reflect the CCD components. Agreement is therefore a measure of prompt adherence, not of fidelity to real client cognition. The evaluation criterion (CCD adherence) is the same construct that was input to the simulator, so the claim that the simulations are 'reliable' and 'accurately emulate' real cognitive processes is a self-consistency check rather than independent evidence.

full rationale

The central claim—that DiaCBT fine-tuning improves CBT-style counseling—is supported by controlled comparisons (Tables 4–5) against baselines in the same simulated environment and by human preference judgments (§5.4). Those comparisons do not reduce to the CCD by construction: no evaluation parameter is fitted to the training data, and the test CCDs are newly generated from C2D2 (§5.1). The genuinely circular element is the validation of the client simulator in §5.5: the simulated client is explicitly prompted with the CCD and told to 'allow your responses to be informed by the diagram,' then experts rate whether the client's utterances reflect the CCD. That confirms prompt compliance, not real-patient fidelity. The paper's own Limitations and Ethical Considerations acknowledge the gap between simulated dialogues and real sessions, which is a validity limitation rather than a circular derivation. Overall circularity is minor.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on domain assumptions about how well LLM-simulated clients and LLM judges capture real therapeutic interaction. The CCD framework itself is borrowed from Beck's CBT literature, and the evaluation's free parameters (epsilon, maximum turns) are not fully specified.

free parameters (2)
  • Success rate threshold epsilon = not specified
    Used in Eq. 2 to determine GOAL-COMPLETED in the LLM reward; the actual value is not reported, so results depend on an unstated choice.
  • Maximum turn count for evaluation = not stated (max allowed turns before termination)
    Success Rate (SR) and Average Turns (AT) are computed within an unspecified turn limit; changing the limit would change SR and AT values.
assumptions (4)
  • domain assumption LLM-simulated clients (GPT-4o) are a valid proxy for real clients in evaluating therapy effectiveness
    Invoked in Section 5.1 and Appendix C.1; without it, reported success rates may not reflect real counseling outcomes.
  • domain assumption Annotated public CBT transcripts are representative of real counseling sessions
    Used as foundation for dialogue generation (Section 3.1); if these transcripts are atypical, generated dialogues inherit bias.
  • domain assumption The six-component CCD fully captures the client's cognitive model for dialogue simulation
    Stated in Section 3.2; if the CCD omits relevant factors, client simulation and evaluation may be incomplete.
  • domain assumption Expert judges' ratings correlate with actual therapeutic success
    Human evaluation (Section 5.4) uses expert comparisons; no outcome-based validity check.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DiaCBT: A Long-Periodic Dialogue Corpus Guided by Cognitive Conceptualization Diagram for CBT-based Psychological Counseling." pith.science (2026). https://pith.science/paper/6PAMUECY

@misc{pith2026250902999,
  author       = {Pith},
  title        = {Pith review of: DiaCBT: A Long-Periodic Dialogue Corpus Guided by Cognitive Conceptualization Diagram for CBT-based Psychological Counseling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6PAMUECY}},
  note         = {Machine review of arXiv:2509.02999}
}
read the original abstract

Psychotherapy reaches only a small fraction of individuals suffering from mental disorders due to social stigma and the limited availability of therapists. Large language models (LLMs), when equipped with professional psychotherapeutic skills, offer a promising solution to expand access to mental health services. However, the lack of psychological conversation datasets presents significant challenges in developing effective psychotherapy-guided conversational agents. In this paper, we construct a long-periodic dialogue corpus for counseling based on cognitive behavioral therapy (CBT). Our curated dataset includes multiple sessions for each counseling and incorporates cognitive conceptualization diagrams (CCDs) to guide client simulation across diverse scenarios. To evaluate the utility of our dataset, we train an in-depth counseling model and present a comprehensive evaluation framework to benchmark it against established psychological criteria for CBT-based counseling. Results demonstrate that DiaCBT effectively enhances LLMs' ability to emulate psychologists with CBT expertise, underscoring its potential for training more professional counseling agents.

Figures

Figures reproduced from arXiv: 2509.02999 by the authors.

Figure 1
Figure 1. A dialogue comparing a previous counseling [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Data creation pipeline of DiaCBT. Step 1: Human annotators collect public cases, annotate CBT sessions, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The distribution of strategies used in conver [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The distribution of ratings. Experts rate over [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: A sample CCD from our DiaCBT. B Details of DiaCBT We provide sample dialogue in [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: The prompt used for DiaCBT to generate dialogues. of interactions that may occur in real-life coun￾seling sessions. This includes evaluating con￾versational flow, tone, and the practical ap￾plication of CBT techniques in an interactive manner. • Situation Diversity: Ex…
Figure 7
Figure 7. Figure 7: The prompt used for DiaCBT to simulate client. vt is not less than a certain threshold ϵ, we re￾gard the state as GOAL-COMPLETED. The map￾ping Mr(·) is a vocabulary such as {"worse": -1.0, "same": -0.5, "better": 0.5, "solved": 1}, the key is the text answer to a multi…
Figure 8
Figure 8. Figure 8: The instructions for annotators to compare the [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. EmoTrace: An Emotion Trajectory-Centered Framework for Psychological Support Dialogue Generation

    cs.CL 2026-07 conditional novelty 6.0 of 10

    A seeker-emotion-trajectory framework with schemas and EFT counselor control yields a 1,114-dialogue corpus and a fine-tuned model that score higher on emotional richness and empathy than prior counseling datasets and bots.

Reference graph

Works this paper leans on

51 extracted references · 39 canonical work pages · cited by 1 Pith paper

  1. [1]

    Gregory A Aarons, Elizabeth A Miller, Amy E Green, Jennifer A Perrott, and Richard Bradway. 2012. Adaptation happens: a qualitative case study of implementation of the incredible years evidence-based parent training programme in a residential substance abuse treatment programme. Journal of Children's Services, 7(4):233--245

    work page 2012

  2. [2]

    Mohammad Rafayet Ali, Seyedeh Zahra Razavi, Raina Langevin, Abdullah Al Mamun, Benjamin Kane, Reza Rawassizadeh, Lenhart K Schubert, and Ehsan Hoque. 2020. A virtual conversational agent for teens with autism spectrum disorder: Experimental results and design lessons. In Proceedings of the 20th ACM international conference on intelligent virtual agents, p...

    work page 2020

  3. [3]

    Judith S Beck. 2020. Cognitive behavior therapy: Basics and beyond

    work page 2020

  4. [4]

    S \'e bastien Bubeck, Varun Chandrasekaran, Ronen Eldan, Johannes Gehrke, Eric Horvitz, Ece Kamar, Peter Lee, Yin Tat Lee, Yuanzhi Li, Scott Lundberg, et al. 2023. Sparks of artificial general intelligence: Early experiments with gpt-4. arXiv preprint arXiv:2303.12712

    arXiv 2023

  5. [5]

    Siyuan Chen, Mengyue Wu, Kenny Q Zhu, Kunyao Lan, Zhiling Zhang, and Lyuchun Cui. 2023 a . Llm-empowered chatbots for psychiatrist and patient simulation: application and evaluation. arXiv preprint arXiv:2305.13614

    arXiv 2023

  6. [6]

    Yirong Chen, Xiaofen Xing, Jingkai Lin, Huimin Zheng, Zhenyu Wang, Qi Liu, and Xiangmin Xu. 2023 b . Soulchat: Improving llms’ empathy, listening, and comfort abilities through fine-tuning with multi-turn empathy conversations. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 1170--1183

    work page 2023

  7. [7]

    Yi Cheng, Wenge Liu, Wenjie Li, Jiashuo Wang, Ruihui Zhao, Bang Liu, Xiaodan Liang, and Yefeng Zheng. 2022. Improving multi-turn emotional support dialogue generation with lookahead strategy planning. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 3014--3026

    work page 2022

  8. [8]

    Yu Ying Chiu, Ashish Sharma, Inna Wanyin Lin, and Tim Althoff. 2024. A computational framework for behavioral assessment of llm therapists. arXiv preprint arXiv:2401.00820

    arXiv 2024

Show all 51 references
  1. [9]

    Berni Curwen, Peter Ruddell, and Stephen Palmer. 2018. Brief cognitive behaviour therapy

    2018

  2. [10]

    Avisha Das, Salih Selek, Alia R Warner, Xu Zuo, Yan Hu, Vipina Kuttichi Keloth, Jianfu Li, W Jim Zheng, and Hua Xu. 2022. Conversational bots for psychotherapy: a study of generative transformer models using domain-specific dialogues. In Proceedings of the 21st Workshop on Bio...

    2022

  3. [11]

    Yang Deng, Wenxuan Zhang, Wai Lam, See-Kiong Ng, and Tat-Seng Chua. 2024. Plug-and-play policy planner for large language model powered dialogue agents

    2024

  4. [12]

    Melvyn Freeman. 2022. The world mental health report: transforming mental health for all. World Psychiatry, 21(3):391

    2022

  5. [13]

    Jun Gao, Yuhan Liu, Haolin Deng, Wei Wang, Yu Cao, Jiachen Du, and Ruifeng Xu. 2021. Improving empathetic response generation by recognizing emotion cause in conversations. In Findings of the association for computational linguistics: EMNLP 2021, pages 807--819

    2021

  6. [14]

    Keith Harrigian, Carlos Aguirre, and Mark Dredze. 2021. On the state of social media data for mental health research. In Proceedings of the Seventh Workshop on Computational Linguistics and Clinical Psychology: Improving Access, pages 15--24

    2021

  7. [15]

    Tao He, Lizi Liao, Yixin Cao, Yuanxing Liu, Ming Liu, Zerui Chen, and Bing Qin. 2024. https://aclanthology.org/2024.acl-long.262 Planning like human: A dual-process framework for dialogue planning . In Proceedings of the 62nd Annual Meeting of the Association for Computational...

    2024

  8. [16]

    Shang-Ling Hsu, Raj Sanjay Shah, Prathik Senthil, Zahra Ashktorab, Casey Dugan, Werner Geyer, and Diyi Yang. 2023. Helping the helper: Supporting peer counselors via ai-empowered practice and feedback. arXiv preprint arXiv:2305.08982

    2023 arXiv

  9. [17]

    Eunkyung Jo, Daniel A Epstein, Hyunhoon Jung, and Young-Ho Kim. 2023. Understanding the benefits and challenges of deploying conversational ai leveraging large language models for public health intervention. In Proceedings of the 2023 CHI Conference on Human Factors in Computi...

    2023

  10. [18]

    Suyeon Lee, Sunghwan Kim, Minju Kim, Dongjin Kang, Dongil Yang, Harim Kim, Minseok Kang, Dayi Jung, Min Hee Kim, Seungbeen Lee, et al. 2024. Cactus: Towards psychological counseling conversations using cognitive behavioral theory. arXiv preprint arXiv:2407.03103

    2024 arXiv

  11. [19]

    Anqi Li, Lizhi Ma, Yaling Mei, Hongliang He, Shuai Zhang, Huachuan Qiu, and Zhenzhong Lan. 2023. https://doi.org/10.18653/v1/2023.acl-long.577 Understanding client reactions in online mental health counseling . In Proceedings of the 61st Annual Meeting of the Association for C...

    2023 doi

  12. [20]

    Qintong Li, Piji Li, Zhaochun Ren, Pengjie Ren, and Zhumin Chen. 2022. Knowledge bridging for empathetic dialogue generation. In Proceedings of the AAAI conference on artificial intelligence, volume 36, pages 10993--11001

    2022

  13. [21]

    June M Liu, Donghao Li, He Cao, Tianhe Ren, Zeyi Liao, and Jiamin Wu. 2023. Chatcounselor: A large language models for mental health support. arXiv preprint arXiv:2309.15461

    2023 arXiv

  14. [22]

    Siyang Liu, Chujie Zheng, Orianna Demasi, Sahand Sabour, Yu Li, Zhou Yu, Yong Jiang, and Minlie Huang. 2021. Towards emotional support dialog systems. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Co...

    2021

  15. [23]

    Nurul Lubis, Sakriani Sakti, Koichiro Yoshino, and Satoshi Nakamura. 2018. Eliciting positive emotion through affect-sensitive dialogue response generation: A neural network approach. In Proceedings of the AAAI conference on artificial intelligence, volume 32

    2018

  16. [24]

    Yukun Ma, Khanh Linh Nguyen, Frank Z Xing, and Erik Cambria. 2020. A survey on empathetic dialogue systems. Information Fusion, 64:50--70

    2020

  17. [25]

    Mounica Maddela, Megan Ung, Jing Xu, Andrea Madotto, Heather Foran, and Y-Lan Boureau. 2023. https://doi.org/10.18653/v1/2023.acl-long.763 Training models to generate, recognize, and reframe unhelpful thoughts . In Proceedings of the 61st Annual Meeting of the Association for ...

    2023 doi

  18. [26]

    Seyed Mahed Mousavi, Alessandra Cervone, Morena Danieli, and Giuseppe Riccardi. 2021. Would you like to tell me more? generating a corpus of psychotherapy dialogues. In Proceedings of the Second Workshop on Natural Language Processing for Medical Conversations, pages 1--9

    2021

  19. [27]

    Hongbin Na. 2024. Cbt-llm: A chinese large language model for cognitive behavioral therapy-based mental health question answering. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), page...

    2024

  20. [28]

    Ver \'o nica P \'e rez-Rosas, Xuetong Sun, Christy Li, Yuchen Wang, Kenneth Resnicow, and Rada Mihalcea. 2018. Analyzing the quality of counseling conversations: the tell-tale signs of high-quality counseling. In Proceedings of the eleventh international conference on language...

    2018

  21. [29]

    Huachuan Qiu, Hongliang He, Shuai Zhang, Anqi Li, and Zhenzhong Lan. 2024 a . https://doi.org/10.18653/v1/2024.findings-emnlp.34 SMILE : Single-turn to multi-turn inclusive language expansion via C hat GPT for mental health support . In Findings of the Association for Computat...

    2024 doi

  22. [30]

    Huachuan Qiu, Anqi Li, Lizhi Ma, and Zhenzhong Lan. 2024 b . Psychat: A client-centric dialogue system for mental health support. In 2024 27th International Conference on Computer Supported Cooperative Work in Design (CSCWD), pages 2979--2984. IEEE

    2024

  23. [31]

    Aravind Sesagiri Raamkumar and Yinping Yang. 2022. Empathetic conversational systems: a review of current advances, gaps, and opportunities. IEEE Transactions on Affective Computing, 14(4):2722--2739

    2022

  24. [32]

    Paolo Raile. 2024. The usefulness of chatgpt for psychotherapists and patients. Humanities and Social Sciences Communications, 11(1):1--8

    2024

  25. [33]

    Hannah Rashkin, Eric Michael Smith, Margaret Li, and Y-Lan Boureau. 2019. Towards empathetic open-domain conversation models: A new benchmark and dataset. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5370--5381

    2019

  26. [34]

    Sahand Sabour, Wen Zhang, Xiyao Xiao, Yuwei Zhang, Yinhe Zheng, Jiaxin Wen, Jialu Zhao, and Minlie Huang. 2023. A chatbot for mental health support: exploring the impact of emohaa on reducing mental distress in china. Frontiers in digital health, 5:1133987

    2023

  27. [35]

    Ashish Sharma, Inna W Lin, Adam S Miner, David C Atkins, and Tim Althoff. 2023. Human--ai collaboration enables more empathic conversations in text-based peer-to-peer mental health support. Nature Machine Intelligence, 5(1):46--57

    2023

  28. [36]

    Ashish Sharma, Adam Miner, David Atkins, and Tim Althoff. 2020. A computational approach to understanding empathy expressed in text-based mental health support. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 5263--5276

    2020

  29. [37]

    Lei Shen, Jinchao Zhang, Jiao Ou, Xiaofang Zhao, and Jie Zhou. 2021. Constructing emotional consensus and utilizing unpaired data for empathetic dialogue generation. In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 3124--3134

    2021

  30. [38]

    Elizabeth C Stade, Shannon Wiltsey Stirman, Lyle H Ungar, Cody L Boland, H Andrew Schwartz, David B Yaden, Jo \ a o Sedoc, Robert J DeRubeis, Robb Willer, and Johannes C Eichstaedt. 2024. Large language models could change the future of behavioral healthcare: a proposal for re...

    2024

  31. [39]

    Bichen Wang, Pengfei Deng, Yanyan Zhao, and Bing Qin. 2023. https://doi.org/10.18653/v1/2023.findings-emnlp.680 C 2 D 2 dataset: A resource for the cognitive distortion analysis and its impact on mental health . In Findings of the Association for Computational Linguistics: EMN...

    2023 doi

  32. [40]

    Chiu, Jiayin Zhi, Shaun M

    Ruiyi Wang, Stephanie Milani, Jamie C. Chiu, Jiayin Zhi, Shaun M. Eack, Travis Labrum, Samuel M Murphy, Nev Jones, Kate V Hardy, Hong Shen, Fei Fang, and Zhiyu Chen. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.711 PATIENT - : Using large language models to simulate patie...

    2024 doi

  33. [41]

    David Watson, Lee Anna Clark, and Auke Tellegen. 1988. Development and validation of brief measures of positive and negative affect: the panas scales. Journal of personality and social psychology, 54(6):1063

    1988

  34. [42]

    Jing Wei, Sungdong Kim, Hyunhoon Jung, and Young-Ho Kim. 2024. Leveraging large language models to power chatbots for collecting user self-reported data. Proceedings of the ACM on Human-Computer Interaction, 8(CSCW1):1--35

    2024

  35. [43]

    Marsha White and Steve M Dorman. 2001. Receiving social support online: implications for health education. Health education research, 16(6):693--707

    2001

  36. [44]

    Zixiu Wu, Simone Balloccu, Vivek Kumar, Rim Helaoui, Ehud Reiter, Diego Reforgiato Recupero, and Daniele Riboni. 2022. Anno-mi: A dataset of expert-annotated counselling dialogues. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (IC...

    2022

  37. [45]

    Mengxi Xiao, Qianqian Xie, Ziyan Kuang, Zhicheng Liu, Kailai Yang, Min Peng, Weiguang Han, and Jimin Huang. 2024. https://aclanthology.org/2024.acl-long.93 H eal M e: Harnessing cognitive reframing in large language models for psychotherapy . In Proceedings of the 62nd Annual ...

    2024

  38. [46]

    An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, et al. 2024. Qwen2. 5 technical report. arXiv preprint arXiv:2412.15115

    2024 arXiv

  39. [47]

    Binwei Yao, Chao Shi, Likai Zou, Lingfeng Dai, Mengyue Wu, Lu Chen, Zhen Wang, and Kai Yu. 2022. D4: a chinese dialogue dataset for depression-diagnosis-oriented chat. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 2438--2459

    2022

  40. [48]

    Chenhao Zhang, Renhao Li, Minghuan Tan, Min Yang, Jingwei Zhu, Di Yang, Jiahao Zhao, Guancheng Ye, Chengming Li, and Xiping Hu. 2024. https://aclanthology.org/2024.findings-acl.830 CP sy C oun: A report-based multi-turn dialogue reconstruction and evaluation framework for C hi...

    2024

  41. [49]

    Hao Zhou, Minlie Huang, Tianyang Zhang, Xiaoyan Zhu, and Bing Liu. 2018. Emotional chatting machine: Emotional conversation generation with internal and external memory. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 32

    2018

  42. [50]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  43. [51]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.