arxiv: 2604.24126 · v1 · submitted 2026-04-27 · 💻 cs.CL

Recognition: unknown

Psychologically-Grounded Graph Modeling for Interpretable Depression Detection

Rishitej Reddy Vyalla , Kritarth Prasad , Avinash Anand , Erik Cambria , Shaoxiong Ji , Faten S. Alamri , Zhengkui Wang

Authors on Pith no claims yet

Pith reviewed 2026-05-08 03:36 UTC · model grok-4.3

classification 💻 cs.CL

keywords depression detectiongraph neural networksconversational analysispsychological modelinginterpretable AIdata augmentationmental health screening

0 comments

The pith

PsyGAT models conversations as dynamic graphs with Psychological Expression Units to detect depression more accurately and with clinical interpretability.

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

The paper introduces PsyGAT to overcome data scarcity and black-box limitations in detecting depression from conversational data. It represents each session as a temporal graph whose nodes are Psychological Expression Units encoding utterance-level clinical evidence, with edges tracking shifts in psychological states. Personality context is embedded in the graph to separate enduring traits from acute symptoms, and persona-based augmentation balances the classes using clinically approved methods. This setup yields higher Macro F1 scores than prior graph models and large language models on standard benchmarks while also improving the ranking of causal symptom triggers. Readers would care because the approach grounds detection in observable psychological patterns, making automated screening both more reliable and more explainable to clinicians.

Core claim

PsyGAT structures conversational sessions as dynamic temporal graphs in which Psychological Expression Units serve as nodes that explicitly encode utterance-level clinical evidence. Graph edges represent transitions between psychological states rather than semantic similarity alone. Session-level personality context is integrated directly into the structure to disentangle trait-based behavior from acute depressive symptoms. Clinically approved persona-based data augmentation addresses class imbalance. The resulting model reaches 89.99 Macro F1 on DAIC-WoZ and 71.37 on E-DAIC, outperforming strong graph baselines and closed-source LLMs such as GPT-5. An attached interpretability module, Causl

What carries the argument

Psychological Expression Units (PEUs), which encode utterance-level clinical evidence to structure temporal graphs that capture transitions in psychological states rather than semantic links.

Load-bearing premise

The newly defined Psychological Expression Units and the persona-based augmentation faithfully encode clinical evidence without introducing dataset-specific biases that inflate the reported performance metrics.

What would settle it

Evaluating the trained PsyGAT model on a fresh, independent set of depression-related conversational recordings drawn from a different clinical source or population to check whether the Macro F1 scores and MRR gains hold without retraining or further augmentation.

Figures

Figures reproduced from arXiv: 2604.24126 by Avinash Anand, Erik Cambria, Faten S. Alamri, Kritarth Prasad, Rishitej Reddy Vyalla, Shaoxiong Ji, Zhengkui Wang.

**Figure 1.** Figure 1: Feature Fusion vs. Psychological Reasoning (left) relies on surface semantics, failing to detect depression in high-functioning individuals who mask their symptoms. In contrast, our approach (right) integrates Persona profiles to interpret text through a clinical lens, extracting Psychological Expression Units (PEUs) that reveal hidden symptoms like Somatic Fatigue and Protective Coping for accurate depr… view at source ↗

**Figure 2.** Figure 2: End-to-end framework for explainable depression detection disentangle sensitive attributes and provide structured explanations for multimodal depression predictions [30]. Despite these advances, most explainable methods rely on attention weights or predefined linguistic features and often operate at the post or user level without explicitly modeling temporal psychological dynamics. This motivates structu… view at source ↗

**Figure 3.** Figure 3: SHAP analysis of individual feature group using without persona (a) and with view at source ↗

**Figure 4.** Figure 4: Evaluate performance when training on different dataset configurations view at source ↗

**Figure 5.** Figure 5: Evaluation on DAIC-WOZ and E-DAIC showing the impact of varying propor view at source ↗

read the original abstract

Automatic depression detection from conversational interactions holds significant promise for scalable screening but remains hindered by severe data scarcity and a lack of clinical interpretability. Existing approaches typically rely on black-box deep learning architectures that struggle to model the subtle, temporal evolution of depressive symptoms or account for participant-specific heterogeneity. In this work, we propose PsyGAT (Psychological Graph Attention Network), a psychologically grounded framework that models conversational sessions as dynamic temporal graphs. We introduce Psychological Expression Units (PEUs) to explicitly encode utterance-level clinical evidence, structuring the session graph to capture transitions in psychological states rather than mere semantic dependencies. To address the critical class imbalance in depression datasets, we employ clinically approved persona-based data augmentation, enable robust model learning. Additionally, we integrate session-level personality context directly into the graph structure to disentangle trait-based behavior from acute depressive symptoms. PsyGAT achieves state-of-the-art performance, surpassing both strong graph-based baselines and closed-source LLMs like GPT-5, achieving 89.99 and 71.37 Macro F1 scores in DAIC-WoZ and E-DAIC, respectively. We further introduce Causal-PsyGAT, an interpretability module that identifies symptom triggers. Experiments show a 20% improvement in MRR for identifying causal indicators, effectively bridging the gap between depression monitoring and clinical explainability. The full augmented dataset is publicly available at https://doi.org/10.6084/m9.figshare.31801921.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

PsyGAT structures conversations as temporal graphs with explicit psychological units and personality context, but the reported SOTA numbers and MRR gains need full experimental controls to hold up.

read the letter

The core of this paper is a graph attention network that turns dialogue sessions into dynamic graphs built around Psychological Expression Units rather than raw semantics, then layers in personality traits and a causal module for spotting symptom triggers. They also release the augmented dataset, which is useful for anyone trying to replicate mental-health NLP work. That combination is the actual novelty: most prior graph models on DAIC-WoZ stay at semantic edges, while this one tries to encode clinical state transitions and trait-versus-state separation directly in the structure. The persona-based augmentation to fix imbalance is a practical move, and claiming gains over both graph baselines and GPT-5 shows they are testing against current strong options. The interpretability angle via Causal-PsyGAT and the 20% MRR lift on causal indicators is the part that could matter for clinical use if it survives scrutiny. The soft spots sit mainly in the validation. The abstract numbers (89.99 and 71.37 Macro F1) look strong, but without seeing the exact baselines, statistical tests, error bars, or ablations on the unaugmented data, it is hard to tell whether the persona augmentation is adding signal or just dataset-specific lexical patterns that the attention layers latch onto. The stress-test worry about non-clinical artifacts is reasonable here; if the paper does not report distribution checks between real and synthetic utterances or run the model on the original imbalanced set, the performance edge stays provisional. The same goes for the GPT-5 comparison: prompting details and temperature settings matter for fair comparison. This work is aimed at NLP researchers who already work on graph models for sequential clinical data and want an interpretable alternative to black-box transformers. A reader who cares about bridging psychology and graph attention will find concrete ideas to build on, even if they end up re-running the experiments themselves. The paper deserves peer review because the framework is coherent and the public data release lowers the barrier to checking the claims, but any referee should press hard on the augmentation controls and the exact experimental protocol before the SOTA label sticks.

Referee Report

3 major / 2 minor

Summary. The paper proposes PsyGAT, a graph attention network that represents conversational sessions as dynamic temporal graphs using newly defined Psychological Expression Units (PEUs) to encode utterance-level clinical evidence for depression detection. It applies persona-based data augmentation to mitigate class imbalance, incorporates session-level personality context into the graph, and introduces Causal-PsyGAT for identifying symptom triggers. The work claims state-of-the-art Macro F1 scores of 89.99 on DAIC-WoZ and 71.37 on E-DAIC (surpassing graph baselines and GPT-5), plus a 20% MRR gain in causal indicator identification, with the augmented dataset released publicly.

Significance. If the empirical claims hold after validation, the integration of psychologically grounded units with temporal graph modeling offers a promising direction for interpretable depression detection, potentially improving clinical adoption by providing both performance and explainability. The public dataset release is a positive contribution to reproducibility in this data-scarce domain.

major comments (3)

[Data Augmentation] Data Augmentation section: The persona-based augmentation is load-bearing for the reported SOTA F1 scores and MRR gains, yet no quantitative validation (such as KL divergence on symptom-transition distributions or ablation comparing augmented vs. unaugmented training) is described to confirm that synthetic utterances preserve real patient distributions rather than introducing exploitable artifacts.
[Experimental Results] Experimental Results section: The headline performance claims (89.99/71.37 Macro F1, 20% MRR improvement) are presented without reference to the full experimental protocol, baseline implementation details, statistical significance tests, error bars across runs, or cross-validation strategy, preventing assessment of whether the gains are robust or dataset-specific.
[Causal-PsyGAT] Causal-PsyGAT subsection: The interpretability module's mechanism for extracting symptom triggers and its validation against clinical standards are not detailed, which is central to the claim of bridging detection performance with explainability.

minor comments (2)

[Abstract] The abstract contains a grammatical fragment ('enable robust model learning') that should be rephrased for clarity.
Notation for PEUs and graph construction could be formalized earlier with an explicit definition or diagram to aid readers unfamiliar with the psychological grounding.

Simulated Author's Rebuttal

3 responses · 0 unresolved

Thank you for the constructive review and the recommendation for major revision. We appreciate the feedback highlighting areas where additional rigor and transparency would strengthen the manuscript. We address each major comment below and commit to incorporating the necessary revisions.

read point-by-point responses

Referee: [Data Augmentation] Data Augmentation section: The persona-based augmentation is load-bearing for the reported SOTA F1 scores and MRR gains, yet no quantitative validation (such as KL divergence on symptom-transition distributions or ablation comparing augmented vs. unaugmented training) is described to confirm that synthetic utterances preserve real patient distributions rather than introducing exploitable artifacts.

Authors: We agree that the manuscript would benefit from explicit quantitative validation of the augmentation. In the revision, we will add an ablation study comparing model performance on original versus augmented training data for both DAIC-WoZ and E-DAIC. We will also compute and report KL divergence on symptom-transition distributions between real and synthetic utterances to demonstrate distributional fidelity. The public release of the augmented dataset already supports external verification of these properties. revision: yes
Referee: [Experimental Results] Experimental Results section: The headline performance claims (89.99/71.37 Macro F1, 20% MRR improvement) are presented without reference to the full experimental protocol, baseline implementation details, statistical significance tests, error bars across runs, or cross-validation strategy, preventing assessment of whether the gains are robust or dataset-specific.

Authors: We acknowledge that greater experimental transparency is required. The revised manuscript will include the complete experimental protocol, detailed baseline implementation descriptions (including any code or hyperparameter references), results with standard error bars from multiple runs, statistical significance testing (e.g., paired t-tests or McNemar's test), and explicit specification of the cross-validation strategy. These additions will allow readers to evaluate the robustness of the reported gains. revision: yes
Referee: [Causal-PsyGAT] Causal-PsyGAT subsection: The interpretability module's mechanism for extracting symptom triggers and its validation against clinical standards are not detailed, which is central to the claim of bridging detection performance with explainability.

Authors: We will substantially expand the Causal-PsyGAT subsection to detail the extraction mechanism, including the precise graph attention and causal inference steps used to identify symptom triggers from the temporal PEU graph. We will also describe the validation procedure, including quantitative comparison to clinical symptom criteria and the exact computation of the reported MRR improvement. This will more clearly link the interpretability module to clinical standards. revision: yes

Circularity Check

0 steps flagged

No significant circularity in claimed derivation chain

full rationale

The paper introduces Psychological Expression Units (PEUs) and persona-based data augmentation as modeling components within the PsyGAT framework, then reports empirical performance on DAIC-WoZ and E-DAIC (89.99/71.37 Macro F1) plus MRR gains for Causal-PsyGAT. No equations, derivations, or self-citations are present that reduce any prediction or result to fitted inputs or prior self-work by construction. The central claims rest on experimental validation of a new graph construction rather than any self-referential loop, making the approach self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 2 invented entities

Abstract supplies no information on free parameters, background axioms, or independent evidence for new entities; ledger entries are therefore empty or minimal.

invented entities (2)

Psychological Expression Units (PEUs) no independent evidence
purpose: Encode utterance-level clinical evidence to structure graphs around psychological-state transitions rather than semantic links.
Newly introduced structuring device described in the abstract.
Causal-PsyGAT no independent evidence
purpose: Interpretability module that identifies symptom triggers and raises MRR for causal indicators.
Additional module proposed for explainability.

pith-pipeline@v0.9.0 · 5585 in / 1357 out tokens · 94385 ms · 2026-05-08T03:36:38.800484+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

37 extracted references · 24 canonical work pages · 1 internal anchor

[1]

B. G. Teferra, A. Perivolaris, W.-N. Hsiang, C. K. Sidharta, A. Rueda, K. Parkington, Y. Wu, A. Soni, R. Samavi, R. Jetly, Y. Zhang, B. Cao, S. Rambhatla, S. Krishnan, V. Bhat, Leveraging large language mod- els for automated depression screening, PLOS Digital Health 4 (2025) e0000943. URL:https://doi.org/10.1371/journal.pdig.0000943. doi:10.1371/journal....

work page doi:10.1371/journal.pdig.0000943 2025
[2]

B. Maji, M. Swain, S. Nasreen, D. Majumdar, R. Guha, A. Routray, A. Søgaard, A study on the impact of foundation models on auto- matic depression detection from speech signals, in: Proceedings of Inter- speech 2025, ISCA, 2025, pp. 5258–5262. doi:10.21437/Interspeech. 2025-1789

work page doi:10.21437/interspeech 2025
[3]

J. M. Liu, M. Gao, S. Sabour, Z. Chen, M. Huang, T. M. C. Lee, Enhanced large language models for effective screening of depression and anxiety, Communications Medicine 5 (2025) 457. URL:https://doi.org/10.1038/s43856-025-01158-1.doi:10.1038/ s43856-025-01158-1

work page doi:10.1038/s43856-025-01158-1.doi:10.1038/ 2025
[4]

Detecting

S. Shreevastava, P. Foltz, Detecting cognitive distortions from patient- therapist interactions, in: N. Goharian, P. Resnik, A. Yates, M. Ireland, K. Niederhoffer, R. Resnik (Eds.), Proceedings of the Seventh Work- shop on Computational Linguistics and Clinical Psychology: Improv- ing Access, Association for Computational Linguistics, Online, 2021, pp.151...

work page doi:10.18653/v1/2021.clpsych-1.17 2021
[5]

LaGrange, D

B. LaGrange, D. Cole, F. Jacquez, J. Ciesla, D. Dallaire, A. Pineda, A. Truss, A. Weitlauf, C. Tilghman-Osborne, J. Felton, Disentangling 26 the prospective relations between maladaptive cognitions and depressive symptoms, JournalofAbnormalPsychology120(2011)511–527.doi:10. 1037/a0024685

2011
[6]

C. Fu, Z. Fu, Q. Zhang, X. Kuang, J. Dong, K. Su, Y. Su, W. Shi, J. Yao, Y. Zhao, S. Zhao, J. Wang, S. Song, C. Liu, Y. Yoshikawa, B. Schuller, H.Ishiguro, Thefirstmpddchallenge: Multimodalpersonality-awarede- pression detection, 2025. URL:https://arxiv.org/abs/2505.10034. arXiv:2505.10034

work page arXiv 2025
[7]

Burdisso, E

S. Burdisso, E. Reyes-Ramírez, E. Villatoro-tello, F. Sánchez-Vega, A. Lopez Monroy, P. Motlicek, DAIC-WOZ: On the validity of using the therapist’s prompts in automatic depression detection from clinical interviews, in: T. Naumann, A. Ben Abacha, S. Bethard, K. Roberts, D. Bitterman (Eds.), Proceedings of the 6th Clinical Natural Language Processing Work...

work page doi:10.18653/v1/2024.clinicalnlp-1.8 2024
[8]

Borsboom, A network theory of mental disorders, World Psychiatry 16 (2017) 5–13

D. Borsboom, A network theory of mental disorders, World Psychiatry 16 (2017) 5–13. doi:10.1002/wps.20375

work page doi:10.1002/wps.20375 2017
[9]

URL: https://www.frontiersin.org/journals/psychology/articles/ 10.3389/fpsyg.2017.01305

P.J.Jones, A.Heeren, R.J.McNally, Commentary: Anetworktheoryof mental disorders, Frontiers in Psychology Volume 8 - 2017 (2017). URL: https://www.frontiersin.org/journals/psychology/articles/ 10.3389/fpsyg.2017.01305. doi:10.3389/fpsyg.2017.01305

work page doi:10.3389/fpsyg.2017.01305 2017
[10]

K.R.Scherer, AppraisalTheory, JohnWiley&Sons, Ltd, 1999, pp.637–

1999
[11]

doi:https://doi.org/10.1002/0470013494.ch30

URL:https://onlinelibrary.wiley.com/doi/abs/10.1002/ 0470013494.ch30. doi:https://doi.org/10.1002/0470013494.ch30. arXiv:https://onlinelibrary.wiley.com/doi/pdf/10.1002/0470013494.ch30

work page doi:10.1002/0470013494.ch30
[12]

2001.Doing Science: Design, Analysis and Communication of Scientific Research(1 ed.)

I. Roseman, C. Smith, Appraisal Theory Overview, Assumptions, Vari- eties, Controversies, 2001, pp. 3–19. doi:10.1093/oso/9780195130072. 003.0001

work page doi:10.1093/oso/9780195130072 2001
[13]

X. Wang, J. Teotia, R. Mao, W. Kaur, E. Cambria, Appraisal theory- informed emotion prediction, in: Proceedings of the International Con- ference on Language Resources and Evaluation (LREC), 2026. 27

2026
[14]

A. T. Beck, Cognitive therapy: Nature and relation to behavior therapy, Behavior therapy 1 (1970) 184–200

1970
[15]

J. P. Allen, An overview of beck ’ s cognitive theory of depression in contemporary literature, 2011. URL:https://api.semanticscholar. org/CorpusID:160019179

2011
[16]

B. Wang, Y. Zhao, X. Lu, B. Qin, Cognitive distortion based explainable depressiondetectionandanalysistechnologiesfortheadolescentinternet users on social media, Frontiers in Public Health 10 (2023) 1045777. doi:10.3389/fpubh.2022.1045777

work page doi:10.3389/fpubh.2022.1045777 2023
[17]

Broerman, Diathesis-Stress Model, Springer International Pub- lishing, Cham, 2017, pp

R. Broerman, Diathesis-Stress Model, Springer International Pub- lishing, Cham, 2017, pp. 1–3. URL:https://doi.org/10.1007/ 978-3-319-28099-8_891-1. doi:10.1007/978-3-319-28099-8_ 891-1

work page doi:10.1007/978-3-319-28099-8_ 2017
[18]

D. N. Klein, R. Kotov, S. J. Bufferd, Personality and depression: ex- planatory models and review of the evidence., Annual review of clini- cal psychology 7 (2011) 269–95. URL:https://api.semanticscholar. org/CorpusID:15886502

2011
[19]

Sadeghi, R

M. Sadeghi, R. Richer, B. Egger, L. Schindler-Gmelch, L. Rupp, F. Rahimi, M. Berking, B. Eskofier, Harnessing multimodal ap- proaches for depression detection using large language models and fa- cial expressions, npj Mental Health Research 3 (2024). doi:10.1038/ s44184-024-00112-8

2024
[20]

P. Cao, Y. Zhang, C. Zhang, W. Chen, Y. Liu, S. Xu, M. Xu, W. Jin, J. Xu, D. Wang, W. Wang, X. Wang, W. Wang, Y. Ren, J. Zhao, R. Li, K. Liu, A multimodal depression consultation dataset of speech and text with hamd-17 assessments, Scientific Data 12 (2025) 1577. URL:https://doi.org/10.1038/s41597-025-05817-9.doi:10.1038/ s41597-025-05817-9

work page doi:10.1038/s41597-025-05817-9.doi:10.1038/ 2025
[21]

Nykoniuk, O

M. Nykoniuk, O. Basystiuk, N. Shakhovska, N. Melnykova, Multimodal data fusion for depression detection approach, Computation 13 (2025)

2025
[22]

doi:10.3390/computation13010009

work page doi:10.3390/computation13010009
[23]

Y. Li, X. Yang, M. Zhao, J. Wang, Y. Yao, W. Qian, S. Qi, Predicting depression by using a novel deep learning model and video-audio-text 28 multimodal data, Frontiers in Psychiatry 16 (2025). URL:https:// api.semanticscholar.org/CorpusID:281546320

2025
[24]

Zhang, J

Y. Zhang, J. You, C. Yang, Z. Wang, X. Qiu, Y. Chen, Deep learning in depression detection: A comprehensive survey and critical analysis, 2025, pp. 177–189. doi:10.1109/ICDEW67478.2025.00028

work page doi:10.1109/icdew67478.2025.00028 2025
[25]

Gomez-Zaragoza, J

L. Gomez-Zaragoza, J. Marin-Morales, M. Alcaniz, M. Soleymani, Speech and text foundation models for depression detection: Cross-task and cross-language evaluation, in: Proceedings of Interspeech 2025, ISCA, 2025, pp. 5253–5257. doi:10.21437/Interspeech.2025-1035

work page doi:10.21437/interspeech.2025-1035 2025
[26]

Zhang, C

X. Zhang, C. Li, W. Chen, J. Zheng, F. Li, Optimizing depression detec- tion in clinical doctor–patient interviews using a multi-instance learning framework, Scientific Reports 15 (2025) 6637. URL:https://doi.org/ 10.1038/s41598-025-90117-w. doi:10.1038/s41598-025-90117-w

work page doi:10.1038/s41598-025-90117-w 2025
[27]

Agarwal, G

N. Agarwal, G. Dias, S. Dollfus, Multi-view graph-based interview rep- resentation to improve depression level estimation, Brain Informatics 11 (2024) 14. URL:https://doi.org/10.1186/s40708-024-00227-w. doi:10.1186/s40708-024-00227-w

work page doi:10.1186/s40708-024-00227-w 2024
[28]

Y. Jiao, K. Zhao, X. Wei, N. B. Carlisle, C. J. Keller, D. J. Oathes, G. A. Fonzo, Y. Zhang, Deep graph learning of multimodal brain networks defines treatment-predictive signatures in major depression, Molecu- lar Psychiatry 30 (2025) 3963–3974. URL:https://doi.org/10.1038/ s41380-025-02974-6. doi:10.1038/s41380-025-02974-6

work page doi:10.1038/s41380-025-02974-6 2025
[29]

L. Zhu, R. Mao, E. Cambria, B. J. Jansen, Neurosymbolic ai for per- sonalized sentiment analysis, in: Proceedings of the 26th International Conference on Human-Computer Interaction (HCII 2024), volume 15382 ofLecture Notes in Computer Science, Springer, 2024, pp. 269–290. doi:10.1007/978-3-031-76827-9_16

work page doi:10.1007/978-3-031-76827-9_16 2024
[30]

G. Tu, B. Wang, E. Cambria, W. Li, R. Xu, Supportplay: A multi- agent role-playing system for personalized and sustained multimodal emotional support conversation, in: Companion Proceedings of the ACM on Web Conference 2025, WWW ’25, Association for Computing Machinery, New York, NY, USA, 2025, p. 2915–2918. URL:https:// doi.org/10.1145/3701716.3715200. d...

work page doi:10.1145/3701716.3715200 2025
[31]

S. Han, R. Mao, E. Cambria, Hierarchical attention network for ex- plainable depression detection on Twitter aided by metaphor concept mappings, in: Proceedings of the 29th International Conference on Computational Linguistics, International Committee on Computational Linguistics, Gyeongju, Republic of Korea, 2022, pp. 94–104. URL: https://aclanthology.or...

2022
[32]

Cheong, S

J. Cheong, S. Kalkan, H. Gunes, Fairrefuse: Referee-guided fusion for multimodal causal fairness in depression detection, in: Proceed- ings of the Thirty-Third International Joint Conference on Artificial Intelligence (IJCAI-24), Special Track on AI for Good, International Joint Conferences on Artificial Intelligence Organization, Macau, China, 2024, pp. ...

2024
[33]

Z. Chen, J. Deng, J. Zhou, J. Wu, T. Qian, M. Huang, Depression detection in clinical interviews with LLM-empowered structural element graph, in: K. Duh, H. Gomez, S. Bethard (Eds.), Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), Associ...

work page doi:10.18653/v1/2024.naacl-long.452 2024
[34]

X. Wang, A. Perez, J. Parapar, F. Crestani, Talkdep: Clinically grounded llm personas for conversation-centric depression screening, CIKM ’25, Association for Computing Machinery, New York, NY, USA, 2025, p. 6554–6558. URL:https://doi.org/10.1145/3746252. 3761617. doi:10.1145/3746252.3761617

work page doi:10.1145/3746252 2025
[35]

I. J. Roseman, C. A. Smith, Appraisal theory: Overview, assumptions, varieties, controversies, in: K. R. Scherer, A. Schorr, T. Johnstone (Eds.), Appraisal Processes in Emotion, Oxford University Press, New York, NY, 2001, pp. 3–19

2001
[36]

K. R. Scherer, Appraisal theory, in: T. Dalgleish, M. J. Power (Eds.), Handbook of Cognition and Emotion, John Wiley & Sons Ltd., Chich- ester, UK, 1999, pp. 637–663. 30

1999
[37]

K. Team, Y. Bai, Y. Bao, Y. Charles, C. Chen, G. Chen, H. Chen, H. Chen, J. Chen, N. Chen, R. Chen, Y. Chen, Y. Chen, Y. Chen, Z. Chen, J. Cui, H. Ding, M. Dong, A. Du, C. Du, D. Du, Y. Du, Y. Fan, Y. Feng, K. Fu, B. Gao, C. Gao, H. Gao, P. Gao, T. Gao, Y. Ge, S. Geng, Q. Gu, X. Gu, L. Guan, H. Guo, J. Guo, X. Hao, T. He, W. He, W. He, Y. He, C. Hong, H. ...

work page internal anchor Pith review arXiv 2026