Pith. sign in

REVIEW 3 major objections 7 minor 54 references

Balancing Knowledge Delivery and Emotional Comfort in Healthcare Conversational Systems

T0 review · 3 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Fine-tuning a medical chatbot on emotionally rewritten doctor-patient dialogues raises its empathy scores while preserving its knowledge-based answers.

desk verdict A useful synthetic-data recipe for empathetic medical dialogues, whose headline claim outruns the evidence because the test set is generated by the same rewriting pipeline that made the training data. read the letter →

arxiv 2506.13692 v1 pith:YZH46AHZ submitted 2025-06-16 cs.CL cs.AI

Shang-Chi Tsai , Yun-Nung Chen This is my paper
classification cs.CLcs.AI
keywords healthcaredialoguesystemsemotionalsupportempathygenerationlargelanguagemodelsdirectpreferenceoptimizationsupervisedfine-tuningmedicalconsultationrewriting
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 asks whether a medical chatbot can comfort a distressed patient and still give accurate medical advice, instead of trading one ability for the other. The authors create training data by having a large language model rewrite real doctor-patient consultations: patient questions are re-expressed with negative emotions such as fear and anxiety, and doctor responses are rewritten to soothe the emotion while keeping the medical content. They fine-tune a 1-billion-parameter instruction-tuned model on this data using three alignment methods—supervised fine-tuning, direct preference optimization, and Kahneman-Tversky optimization—and evaluate on held-out emotional consultations. Their central claim is that the fine-tuned models score higher on emotion-focused evaluation than the base model and a prompt-only baseline, while lexical similarity metrics show the medical knowledge is retained.

What carries the argument

The load-bearing mechanism is the synthetic emotional-dialogue rewrite pipeline combined with preference fine-tuning. The central object is the Emotional Question plus Soothing Response subset, in which patient utterances are rewritten to carry one of five negative emotions and doctor responses are rewritten to soothe that emotion without dropping the medical content. Direct preference optimization treats the rewritten soothing response as the preferred completion and the original terse doctor reply as the rejected completion, which is what pushes the model toward emotional phrasing; supervised fine-tuning and KTO absorb the same data through different losses. The Empathetic Response subset supplies auxiliary empathetic style for ordinary questions.

What would settle it

Collect naturally occurring, un-rewritten expressions of patient distress and have real patients or clinicians compare the fine-tuned model's responses with the base model on both comfort and medical usefulness; if the empathy advantage disappears or the soothing responses contain more factual omissions on those naturally occurring inputs, the central claim fails.

Watch

Extended reading notes

Core claim

The discovery is that emotional competence in medical dialogue can be instilled through rewriting rather than by collecting new clinician-patient data. From real consultations, the paper builds two kinds of altered dialogues: Empathetic Responses, where ordinary doctor replies are made warmer, and Emotional Question plus Soothing Response dialogues, where patient messages are made explicitly fearful, anxious, embarrassed, frustrated, or distrustful and the doctor response is rewritten to acknowledge the feeling first and then deliver the medical information. Fine-tuning on these dialogues, especially on the Emotional Question plus Soothing Response subset with direct preference optimization, raises the mean emotion score from 0.54 on the base model to 0.67 as judged by an emotion-specialised evaluator, while the supervised and KTO variants produce the largest gains in BLEU and ROUGE overlap with reference responses. The paper's claim is that a model can learn both roles at once: lead with empathy, then provide knowledge.

Load-bearing premise

The method assumes that dialogues rewritten by an LLM to contain scripted negative emotions stand in for how real distressed patients speak, so a model trained on them will comfort real patients rather than merely imitate the rewriting style.

Editorial extensions

If this is right

  • A medical chatbot trained this way tends to open with emotional acknowledgment and then give medical advice, as shown in the paper's qualitative examples.
  • Direct preference optimization on the Emotional Question plus Soothing Response data yields the highest emotional-intensity scores with a smaller training subset than combining both datasets.
  • Supervised fine-tuning and KTO give the strongest BLEU and ROUGE agreement with reference doctor responses, indicating that emotional style does not necessarily erase factual content.
  • Pre-fine-tuning on Empathetic Response data makes model outputs preferred by an LLM judge on both knowledge and emotional dimensions, regardless of the later alignment method.
  • The recipe is demonstrated on a 1-billion-parameter model trained on a single 32GB GPU, so the approach is practical to reproduce without large-scale compute.

Reading between the lines

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

  • The leading open question is whether LLM-rewritten negative emotions transfer to how real patients actually express distress; until a human test is run, the emotion-score gains may partly reflect the rewriting distribution rather than genuine comfort.
  • An LLM-based emotion evaluator may reward formulaic empathetic phrasing, so a clinician- or patient-preference study would be a stronger test of whether the soothing responses are actually reassuring.
  • The single-turn design leaves untested how the approach behaves over a multi-turn consultation, where a patient's emotional state evolves and trust builds across several exchanges.
  • The paper's own qualitative note that DPO can de-emphasise knowledge suggests combining preference fine-tuning with knowledge-grounded retrieval or a factual-coherence constraint as a next step.
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

3 major / 7 minor

Summary. The paper proposes a pipeline for emotionally aware medical dialogue: an LLM rewrites a real-world doctor-patient dialogue dataset to inject negative patient emotions (fear, anxiety, embarrassment, frustration, distrust) and to produce soothing but knowledge-preserving doctor responses, creating ER and EQ+SR subsets. The authors fine-tune LLaMA-3.2-1B with SFT, DPO, or KTO on these data, then evaluate with EmoLLaMA emotion scores, BLEU/ROUGE against original and modified responses, and Qwen-based preference selections. The paper claims the fine-tuned models significantly improve emotional expression while retaining accurate medical knowledge.

Significance. If the evaluation were trustworthy, the contribution would be useful: the paper releases code, compares three preference-tuning methods in a medical setting, and constructs a large synthetic emotional dialogue resource. The practical finding that DPO improves emotional phrasing while SFT/KTO achieve higher n-gram similarity is potentially informative. However, the current measurements do not establish transfer to real patients or preservation of medical safety, so the significance remains contingent on stronger evaluation.

major comments (3)
  1. [Section 4.1 (Setup)] The test set is 'adapted using the same methodology as the EQ+SR subset of the training set,' which means the same gpt-4o-mini prompt family from Section 3.1 produced both the training and the evaluation patient questions and doctor responses. As a result, Tables 1 and 2 measure performance on the synthetic rewriting distribution rather than on genuine emotional consultations. The abstract's claim that the method helps with real patient interactions is not supported unless an external test set or independently collected emotional dialogues is used.
  2. [Section 4.2 and Table 1] Emotional quality is measured by a single LLM judge, EmoLLaMA, with no human validation, no inter-annotator agreement, and no error bars or significance tests. The reported gains (e.g., mean emotion score from 0.54 to 0.67) are small and may reflect the judge's preference for formulaic empathy rather than appropriate comfort. The Qwen-based preference selections in Figure 2 are also unvalidated LLM judgments. A human or preference-based evaluation with significance testing is needed to support the 'significantly enhances' claim.
  3. [Section 4.3.2 and Table 2] Knowledge retention is quantified only by BLEU/ROUGE overlap with the original and the LLM-modified doctor responses. N-gram similarity is not medical correctness, and the 'Modified response as label' columns compare against the same rewritten responses used as training targets, so high overlap partly measures imitation of the training distribution. The qualitative example in Table 3 (case (a)) shows the DPO model stating 'I'm also going to prescribe you some antiretroviral medication' without any indication or risk assessment, which word-overlap metrics would not penalize. An evaluation by medical experts or a factual-consistency metric is required to support the claim that 'accurate knowledge-based answers' are maintained.
minor comments (7)
  1. [Section 4.1] The dataset description is arithmetically inconsistent: the text says the training set has 100k entries, but then splits it into approximately 60k ER entries and 50k EQ+SR entries, totaling roughly 110k. Please clarify the exact split or the degree of overlap.
  2. [Section 3.1] There are typos in the data description and prompt template: 'Emotional Question(RQ)' should be 'Emotional Question (EQ)', and 'a patient and a dotor' should be 'a patient and a doctor'.
  3. [Section 4.3.1] The sentence 'fine-tuning with EQ+SR data using DPO improved the average and maximum metrics by 0.03 and 0.13, respectively' does not clearly match Table 1; please verify the exact comparison points (base model or prompt-based approach) and the reported differences.
  4. [Equation (5)] The notation 'Regex(y,y*_x)' in the KTO loss is unclear; please define the indicator function and avoid using a string-matching term for what appears to be a desirability label.
  5. [Introduction and References] The citation '(Tsai et al.; Angel et al.)' in the Introduction is incomplete and not properly formatted; please provide full references or remove the placeholder.
  6. [Section 4.2 and Table 1] The name of the emotion evaluator is inconsistent ('EmoLLaMA' vs. 'Emollama-chat-7b'); please standardize the spelling throughout the paper.
  7. [Conclusion and Limitations] Given the medical domain and the unsafe example in Table 3, the paper should include a limitations and safety paragraph discussing the need for clinician oversight and the risks of automated prescribing language.

Circularity Check

2 steps flagged · score 6.0 of 10

The central evaluation is partially circular: the test set is adapted with the same LLM-rewriting pipeline that generated the training data, and the 'Modified response as label' scores reward exactly the responses the models were trained to produce, though comparisons against original doctor responses and among fine-tuning methods retain independent content.

  1. fitted input called prediction [Section 4.1 (Setup).]
    "To evaluate the fine-tuned models, we measured accuracy on a test set adapted using the same methodology as the EQ+SR subset of the training set. This ensured consistency in assessing the model's ability to address queries expressing negative emotions and provide corresponding alleviating responses."

    The test set is not independent: it is produced by the same LLM-rewriting pipeline (gpt-4o-mini with the Section 3.1 prompt template) that created the EQ+SR training data. The models are trained to maximize the likelihood of these rewritten responses (SFT loss, Eq. 1) or to prefer them over original responses (DPO loss, Eq. 2). Therefore, the reported emotion and knowledge scores on this adapted test set measure reproduction of the training distribution, not generalization to real emotional consultations; the claimed prediction is effectively scored on its own training generator.

  2. fitted input called prediction [Section 3.3 (DPO) and Table 2.]
    "To convert the dataset we constructed in Section 3.1 into the format required for DPO, we treated the modified soothing responses as the preferred responses and the original doctor responses as the rejected responses."

    The training objective explicitly increases the likelihood of the modified soothing responses over the original responses, and Table 2's 'Modified response as label' rows use exactly those modified responses as ground truth. The large BLEU/ROUGE gains against the modified labels are therefore entailed by the training setup: the model was optimized to output those labels. These gains are not independent evidence of knowledge preservation. The comparison against original doctor responses is less forced, but it still relies on n-gram overlap rather than medical accuracy.

full rationale

The paper's fine-tuning recipe is self-consistent: it creates a synthetic EQ+SR dataset by prompting gpt-4o-mini to inject negative emotions and soothing responses, trains models to imitate those rewrites (SFT loss Eq. 1; DPO/KTO preference pairs in Sections 3.3-3.4), and then evaluates on a test set 'adapted using the same methodology as the EQ+SR subset of the training set' (Section 4.1). This makes the central emotion and knowledge evaluation partly in-distribution: the model is being scored on the same rewriting distribution it was optimized to reproduce. The most concrete reduction is Table 2's 'Modified response as label' column: the ground-truth labels are exactly the modified soothing responses that SFT maximizes the likelihood of and that DPO/KTO set as preferred responses, so the large BLEU/ROUGE gains against those labels are entailed by the training objective rather than independently demonstrating knowledge preservation. The same-methodology test set also weakens the abstract's claim of improved emotional responses as evidence of transfer to real patient consultations; Section 5's statement that the model was 'tested on real-world emotional conversations' overstates this, since the test set was itself rewritten. The comparison against original doctor responses (Table 2, top) and the relative ranking of SFT/DPO/KTO (Table 1, Figure 2) retain independent content, so the circularity is partial rather than total. No load-bearing self-citation or imported uniqueness theorem was found.

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

The paper introduces no novel mathematical objects, forces, or entities. The main assumptions are about the validity of LLM-based rewriting for creating training data, the reliability of LLM-based emotion scoring, and the adequacy of n-gram metrics for knowledge retention. Two hand-chosen design parameters (the data split and the emotion categories) are not fitted but are selected without explicit justification.

free parameters (2)
  • ER/EQ+SR data split ratio = 60k / 50k
    The training set was divided into approximately 60k empathetic-response entries and 50k emotional-question entries without a stated principled criterion. This balance affects how strongly the model learns each objective.
  • Negative emotion categories = fear, anxiety, embarrassment, frustration, distrust
    Five emotions were selected by hand for the EQ+SR rewriting prompts; the set is not justified and may not cover the full range of negative emotions patients express.
assumptions (4)
  • domain assumption The LLM-rewritten dialogues preserve the original medical information while adding the target emotional tone.
    The data modification in Section 3.1 relies on gpt-4o-mini and LLaMA-3.1 to rewrite both patient queries and doctor responses without altering medical content. No manual verification or quality metric for information preservation is reported.
  • domain assumption EmoLLaMA's emotion intensity scores are a valid measure of empathetic, comforting, and reassuring content.
    The emotion scores in Table 1 are the primary evidence for improved emotional support, but the scores come from a single LLM annotator without validation against human judgments.
  • domain assumption ROUGE/BLEU overlap with the original doctor responses is a valid proxy for medical knowledge retention.
    Section 4.3.2 uses n-gram metrics against the original doctor responses as the knowledge score; lexical overlap is an imperfect measure of medical correctness.
  • domain assumption Single-turn utterances are sufficient to represent the consultation scenarios.
    All dialogues were processed as single-turn utterances (Section 3.1), which simplifies the problem but may not capture the multi-turn emotional dynamics of real consultations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Balancing Knowledge Delivery and Emotional Comfort in Healthcare Conversational Systems." pith.science (2026). https://pith.science/paper/YZH46AHZ

@misc{pith2026250613692,
  author       = {Pith},
  title        = {Pith review of: Balancing Knowledge Delivery and Emotional Comfort in Healthcare Conversational Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YZH46AHZ}},
  note         = {Machine review of arXiv:2506.13692}
}
read the original abstract

With the advancement of large language models, many dialogue systems are now capable of providing reasonable and informative responses to patients' medical conditions. However, when patients consult their doctor, they may experience negative emotions due to the severity and urgency of their situation. If the model can provide appropriate comfort and empathy based on the patient's negative emotions while answering medical questions, it will likely offer a more reassuring experience during the medical consultation process. To address this issue, our paper explores the balance between knowledge sharing and emotional support in the healthcare dialogue process. We utilize a large language model to rewrite a real-world interactive medical dialogue dataset, generating patient queries with negative emotions and corresponding medical responses aimed at soothing the patient's emotions while addressing their concerns. The modified data serves to refine the latest large language models with various fine-tuning methods, enabling them to accurately provide sentences with both emotional reassurance and constructive suggestions in response to patients' questions. Compared to the original LLM model, our experimental results demonstrate that our methodology significantly enhances the model's ability to generate emotional responses while maintaining its original capability to provide accurate knowledge-based answers.

Figures

Figures reproduced from arXiv: 2506.13692 by the authors.

Figure 1
Figure 1. Illustration of an example about the emotional healthcare dialogue system. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Preference selection based on the knowledgeable and emotional dimensions of Qwen’s responses. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

54 extracted references · 15 canonical work pages

  1. [1]

    Asmr: Augmenting life scenario using large generative models for robotic action reflection

    Shang-Chi Tsai Seiya Kawano Angel, Garcia Contreras Koichiro Yoshino, and Yun-Nung Chen. Asmr: Augmenting life scenario using large generative models for robotic action reflection

    work page

  2. [2]

    Yirong Chen, Zhenyu Wang, Xiaofen Xing, huimin zheng, Zhipei Xu, Kai Fang, Junhong Wang, Sihang Li, Jieling Wu, Qi Liu, and Xiangmin Xu. 2023 a . https://arxiv.org/abs/2310.15896 Bianque: Balancing the questioning and suggestion ability of health llms with multi-turn health conversations polished by chatgpt . Preprint, arXiv:2310.15896

    arXiv 2023

  3. [3]

    Zeming Chen, Alejandro Hernández Cano, Angelika Romanou, Antoine Bonnet, Kyle Matoba, Francesco Salvi, Matteo Pagliardini, Simin Fan, Andreas Köpf, Amirkeivan Mohtashami, Alexandre Sallinen, Alireza Sakhaeirad, Vinitra Swamy, Igor Krawczuk, Deniz Bayazit, Axel Marmet, Syrielle Montariol, Mary-Anne Hartley, Martin Jaggi, and Antoine Bosselut. 2023 b . http...

    arXiv 2023

  4. [4]

    Kawin Ethayarajh, Winnie Xu, Niklas Muennighoff, Dan Jurafsky, and Douwe Kiela. 2024. https://arxiv.org/abs/2402.01306 Kto: Model alignment as prospect theoretic optimization . Preprint, arXiv:2402.01306

    arXiv 2024

  5. [5]

    Churpek, and Majid Afshar

    Yanjun Gao, Ruizhe Li, John Caskey, Dmitriy Dligach, Timothy Miller, Matthew M. Churpek, and Majid Afshar. 2023. https://arxiv.org/abs/2308.14321 Leveraging a medical knowledge graph into large language models for diagnosis prediction . Preprint, arXiv:2308.14321

    arXiv 2023

  6. [6]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, Arun Rao, Aston Zhang, Aurelien Rodriguez, Austen Gregerson, Ava S...

    arXiv 2024

  7. [7]

    Adams, Jens-Michalis Papaioannou, Paul Grundmann, Tom Oberhauser, Alexander Löser, Daniel Truhn, and Keno K

    Tianyu Han, Lisa C. Adams, Jens-Michalis Papaioannou, Paul Grundmann, Tom Oberhauser, Alexander Löser, Daniel Truhn, and Keno K. Bressem. 2023. https://arxiv.org/abs/2304.08247 Medalpaca -- an open-source collection of medical conversational ai models and training data . Preprint, arXiv:2304.08247

    arXiv 2023

  8. [8]

    Chao-Wei Huang, Shang-Chi Tsai, and Yun-Nung Chen. 2022. https://doi.org/10.18653/v1/2022.clinicalnlp-1.2 PLM - ICD : Automatic ICD coding with pretrained language models . In Proceedings of the 4th Clinical Natural Language Processing Workshop, pages 10--20, Seattle, WA. Association for Computational Linguistics

    doi:10.18653/v1/2022.clinicalnlp-1.2 2022

Show all 54 references
  1. [9]

    Cheng Li, Jindong Wang, Yixuan Zhang, Kaijie Zhu, Wenxin Hou, Jianxun Lian, Fang Luo, Qiang Yang, and Xing Xie. 2023 a . https://arxiv.org/abs/2307.11760 Large language models understand and can be enhanced by emotional stimuli . Preprint, arXiv:2307.11760

    2023 arXiv

  2. [10]

    Li, Xiaodan Liang, Zhiting Hu, and Eric P

    Christy Y. Li, Xiaodan Liang, Zhiting Hu, and Eric P. Xing. 2018. https://arxiv.org/abs/1805.08298 Hybrid retrieval-generation reinforced agent for medical image report generation . Preprint, arXiv:1805.08298

    2018 arXiv

  3. [11]

    Yunxiang Li, Zihan Li, Kai Zhang, Ruilong Dan, Steve Jiang, and You Zhang. 2023 b . Chatdoctor: A medical chat model fine-tuned on a large language model meta-ai (llama) using medical domain knowledge. Cureus, 15(6)

    2023

  4. [12]

    Kangenbei Liao, CHENG ZHONG, Wei Chen, Qianlong Liu, zhongyu wei, Baolin Peng, and Xuanjing Huang. 2022. https://openreview.net/forum?id=8kVP8m93VqN Task-oriented dialogue system for automatic disease diagnosis via hierarchical reinforcement learning

    2022

  5. [13]

    Chin-Yew Lin. 2004. https://aclanthology.org/W04-1013/ ROUGE : A package for automatic evaluation of summaries . In Text Summarization Branches Out, pages 74--81, Barcelona, Spain. Association for Computational Linguistics

    2004

  6. [14]

    Zhengliang Liu, Yiwei Li, Peng Shu, Aoxiao Zhong, Longtao Yang, Chao Ju, Zihao Wu, Chong Ma, Jie Luo, Cheng Chen, Sekeun Kim, Jiang Hu, Haixing Dai, Lin Zhao, Dajiang Zhu, Jun Liu, Wei Liu, Dinggang Shen, Tianming Liu, Quanzheng Li, and Xiang Li. 2023. https://arxiv.org/abs/23...

    2023 arXiv

  7. [15]

    Zhiwei Liu, Kailai Yang, Tianlin Zhang, Qianqian Xie, Zeping Yu, and Sophia Ananiadou. 2024. Emollms: A series of emotional large language models and annotation tools for comprehensive affective analysis. arXiv preprint arXiv:2401.08508

    2024 arXiv

  8. [16]

    Usman Naseem, Ajay Bandi, Shaina Raza, Junaid Rashid, and Bharathi Raja Chakravarthi. 2022. https://doi.org/10.18653/v1/2022.bionlp-1.10 Incorporating medical knowledge to transformer-based language models for medical dialogue generation . In Proceedings of the 21st Workshop o...

    2022 doi

  9. [17]

    OpenAI, :, Aaron Hurst, Adam Lerer, Adam P. Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, Aleksander Mądry, Alex Baker-Whitcomb, Alex Beutel, Alex Borzunov, Alex Carney, Alex Chow, Alex Kirillov, Alex Nichol, Alex Pai...

    2024 arXiv

  10. [18]

    OpenAI. 2023. https://arxiv.org/abs/2303.08774 Gpt-4 technical report . Preprint, arXiv:2303.08774

    2023 arXiv

  11. [19]

    Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and...

    2022 arXiv

  12. [20]

    Samuel J. Paech. 2024. https://arxiv.org/abs/2312.06281 Eq-bench: An emotional intelligence benchmark for large language models . Preprint, arXiv:2312.06281

    2024 arXiv

  13. [21]

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. https://doi.org/10.3115/1073083.1073135 Bleu: a method for automatic evaluation of machine translation . In Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, ACL '02, page 311...

    2002

  14. [22]

    Manning, and Chelsea Finn

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, and Chelsea Finn. 2024. https://arxiv.org/abs/2305.18290 Direct preference optimization: Your language model is secretly a reward model . Preprint, arXiv:2305.18290

    2024 arXiv

  15. [23]

    Sahand Sabour, Siyang Liu, Zheyuan Zhang, June Liu, Jinfeng Zhou, Alvionna Sunaryo, Tatia Lee, Rada Mihalcea, and Minlie Huang. 2024. https://doi.org/10.18653/v1/2024.acl-long.326 E mo B ench: Evaluating the emotional intelligence of large language models . In Proceedings of t...

    2024 doi

  16. [24]

    Xiaoming Shi, Zeming Liu, Li Du, Yuxuan Wang, Hongru Wang, Yuhang Guo, Tong Ruan, Jie Xu, Xiaofan Zhang, and Shaoting Zhang. 2024. https://doi.org/10.18653/v1/2024.findings-acl.167 Medical dialogue system: A survey of categories, methods, evaluation and challenges . In Finding...

    2024 doi

  17. [25]

    Karan Singhal, Shekoofeh Azizi, Tao Tu, S. Sara Mahdavi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tanwani, Heather Cole-Lewis, Stephen Pfohl, Perry Payne, Martin Seneviratne, Paul Gamble, Chris Kelly, Nathaneal Scharli, Aakanksha Chowdhery, Philip Mansfield, Blaise Ague...

    2022 arXiv

  18. [26]

    Sara Mahdavi, Joelle Barral, Dale Webster, Greg S

    Karan Singhal, Tao Tu, Juraj Gottweis, Rory Sayres, Ellery Wulczyn, Le Hou, Kevin Clark, Stephen Pfohl, Heather Cole-Lewis, Darlene Neal, Mike Schaekermann, Amy Wang, Mohamed Amin, Sami Lachgar, Philip Mansfield, Sushant Prakash, Bradley Green, Ewa Dominowska, Blaise Aguera y ...

    2023 arXiv

  19. [27]

    Kai-Fu Tang. 2016. https://api.semanticscholar.org/CorpusID:35812678 Inquire and diagnose : Neural symptom checking ensemble using deep reinforcement learning

    2016

  20. [28]

    Xiangru Tang, Anni Zou, Zhuosheng Zhang, Ziming Li, Yilun Zhao, Xingyao Zhang, Arman Cohan, and Mark Gerstein. 2024. https://arxiv.org/abs/2311.10537 Medagents: Large language models as collaborators for zero-shot medical reasoning . Preprint, arXiv:2311.10537

    2024 arXiv

  21. [29]

    Chongyang Tao, Jiazhan Feng, Chang Liu, Juntao Li, Xiubo Geng, and Daxin Jiang. 2021. https://arxiv.org/abs/2110.00159 Building an efficient and effective retrieval-based dialogue system via mutual learning . Preprint, arXiv:2110.00159

    2021 arXiv

  22. [30]

    Lawler, Jimmy Ba, Rahul G

    Augustin Toma, Patrick R. Lawler, Jimmy Ba, Rahul G. Krishnan, Barry B. Rubin, and Bo Wang. 2023. https://arxiv.org/abs/2305.12031 Clinical camel: An open expert-level medical language model with dialogue-based knowledge encoding . Preprint, arXiv:2305.12031

    2023 arXiv

  23. [31]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023 a . https://arxiv.org/abs/2302.13971 Lla...

    2023 arXiv

  24. [32]

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, W...

    2023 arXiv

  25. [33]

    https://api.semanticscholar.org/CorpusID:268043378 Asmr: Augmenting life scenario using large generative models for robotic action reflection

    Shang-Chi Tsai, Seiya Kawano, Angel Fernando Garcia Contreras, Koichiro Yoshino, and Yun-Nung Chen. https://api.semanticscholar.org/CorpusID:268043378 Asmr: Augmenting life scenario using large generative models for robotic action reflection

  26. [34]

    Guangyu Wang, Guoxing Yang, Zongxin Du, Longjun Fan, and Xiaohu Li. 2023 a . https://arxiv.org/abs/2306.09968 Clinicalgpt: Large language models finetuned with diverse medical data and comprehensive evaluation . Preprint, arXiv:2306.09968

    2023 arXiv

  27. [35]

    Haochun Wang, Chi Liu, Nuwa Xi, Zewen Qiang, Sendong Zhao, Bing Qin, and Ting Liu. 2023 b . https://arxiv.org/abs/2304.06975 Huatuo: Tuning llama model with chinese medical knowledge . Preprint, arXiv:2304.06975

    2023 arXiv

  28. [36]

    Hongru Wang, Lingzhi Wang, Yiming Du, Liang Chen, Jingyan Zhou, Yufei Wang, and Kam-Fai Wong. 2023 c . https://arxiv.org/abs/2311.16789 A survey of the evolution of language model-based dialogue systems . Preprint, arXiv:2311.16789

    2023 arXiv

  29. [37]

    Sheng Wang, Zihao Zhao, Xi Ouyang, Qian Wang, and Dinggang Shen. 2023 d . https://arxiv.org/abs/2302.07257 Chatcad: Interactive computer-aided diagnosis on medical image using large language models . Preprint, arXiv:2302.07257

    2023 arXiv

  30. [38]

    Xuena Wang, Xueting Li, Zi Yin, Yue Wu, and Liu Jia. 2023 e . https://arxiv.org/abs/2307.09042 Emotional intelligence of large language models . Preprint, arXiv:2307.09042

    2023 arXiv

  31. [39]

    Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M

    Jason Wei, Maarten Bosma, Vincent Y. Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M. Dai, and Quoc V. Le. 2022. https://arxiv.org/abs/2109.01652 Finetuned language models are zero-shot learners . Preprint, arXiv:2109.01652

    2022 arXiv

  32. [40]

    Zhongyu Wei, Qianlong Liu, Baolin Peng, Huaixiao Tou, Ting Chen, Xuanjing Huang, Kam-fai Wong, and Xiangying Dai. 2018. https://doi.org/10.18653/v1/P18-2033 Task-oriented dialogue system for automatic diagnosis . In Proceedings of the 56th Annual Meeting of the Association for...

    2018 doi

  33. [41]

    Chaoyi Wu, Weixiong Lin, Xiaoman Zhang, Ya Zhang, Yanfeng Wang, and Weidi Xie. 2023. https://arxiv.org/abs/2304.14454 Pmc-llama: Towards building open-source language models for medicine . Preprint, arXiv:2304.14454

    2023 arXiv

  34. [42]

    Honglin Xiong, Sheng Wang, Yitao Zhu, Zihao Zhao, Yuxiao Liu, Linlin Huang, Qian Wang, and Dinggang Shen. 2023. https://arxiv.org/abs/2304.01097 Doctorglm: Fine-tuning your chinese doctor is not a herculean task . Preprint, arXiv:2304.01097

    2023 arXiv

  35. [43]

    Dey, and Dakuo Wang

    Xuhai Xu, Bingsheng Yao, Yuanzhe Dong, Saadia Gabriel, Hong Yu, James Hendler, Marzyeh Ghassemi, Anind K. Dey, and Dakuo Wang. 2024. https://doi.org/10.1145/3643540 Mental-llm: Leveraging large language models for mental health prediction via online text data . Proceedings of ...

    2024 doi

  36. [44]

    An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jin Xu, Jingren Zhou, Jinze Bai, Jinzheng...

    2024 arXiv

  37. [45]

    An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, Le Yu, Me...

    2024 arXiv

  38. [46]

    Xingyi Yang, Muchao Ye, Quanzeng You, and Fenglong Ma. 2021. https://doi.org/10.18653/v1/2021.acl-long.387 Writing by memorizing: Hierarchical retrieval-based medical report generation . In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics...

    2021 doi

  39. [47]

    Qichen Ye, Junling Liu, Dading Chong, Peilin Zhou, Yining Hua, Fenglin Liu, Meng Cao, Ziming Wang, Xuxin Cheng, Zhu Lei, and Zhenhua Guo. 2024. https://arxiv.org/abs/2310.09089 Qilin-med: Multi-stage knowledge injection advanced medical large language model . Preprint, arXiv:2...

    2024 arXiv

  40. [48]

    Shengyu Zhang, Linfeng Dong, Xiaoya Li, Sen Zhang, Xiaofei Sun, Shuhe Wang, Jiwei Li, Runyi Hu, Tianwei Zhang, Fei Wu, and Guoyin Wang. 2024. https://arxiv.org/abs/2308.10792 Instruction tuning for large language models: A survey . Preprint, arXiv:2308.10792

    2024

  41. [49]

    Zheng Zhang, Ryuichi Takanobu, Qi Zhu, Minlie Huang, and Xiaoyan Zhu. 2020. https://arxiv.org/abs/2003.07490 Recent advances and challenges in task-oriented dialog system . Preprint, arXiv:2003.07490

    2020 arXiv

  42. [50]

    Yu Zhao, Yunxin Li, Yuxiang Wu, Baotian Hu, Qingcai Chen, Xiaolong Wang, Yuxin Ding, and Min Zhang. 2022. https://doi.org/10.1145/3534678.3542674 Medical dialogue response generation with pivotal information recalling . In Proceedings of the 28th ACM SIGKDD Conference on Knowl...

    2022

  43. [51]

    Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, Zhangchi Feng, and Yongqiang Ma. 2024. https://arxiv.org/abs/2403.13372 Llamafactory: Unified efficient fine-tuning of 100+ language models . Preprint, arXiv:2403.13372

    2024 arXiv

  44. [52]

    Cheng Zhong, Kangenbei Liao, Wei Chen, Qianlong Liu, Baolin Peng, Xuanjing Huang, Jiajie Peng, and Zhongyu Wei. 2023. https://arxiv.org/abs/2004.14254 Hierarchical reinforcement learning for automatic disease diagnosis . Preprint, arXiv:2004.14254

    2023 arXiv

  45. [53]

    Meng Zhou, Zechen Li, Bowen Tan, Guangtao Zeng, Wenmian Yang, Xuehai He, Zeqian Ju, Subrato Chakravorty, Shu Chen, Xingyi Yang, Yichen Zhang, Qingyang Wu, Zhou Yu, Kun Xu, Eric Xing, and Pengtao Xie. 2021. https://doi.org/10.18653/v1/2021.acl-short.112 On the generation of med...

    2021 doi

  46. [54]

    Ying Zhu, Shi Feng, Daling Wang, Yifei Zhang, and Donghong Han. 2022. Knowledge-enhanced interactive matching network for multi-turn response selection in medical dialogue systems. In Database Systems for Advanced Applications, pages 255--262, Cham. Springer International Publishing

    2022

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.