pith. sign in

arxiv: 2606.23387 · v1 · pith:5JJHZ6BOnew · submitted 2026-06-22 · 💻 cs.CL

Self-Stigma Is Not a Monolith, but Generic Empathy Is: Persona-Conditioned LLM Support for People Who Use Drugs

Layla Bouzoubaa , Rezvaneh Rezapour This is my paper

Pith reviewed 2026-06-26 08:04 UTC · model grok-4.3

classification 💻 cs.CL
keywords self-stigmaLLM supportpersona classificationpeople who use drugsempathy evaluationconversational AIReddit analysislatent profile analysis
0
0 comments X

The pith

Persona-matched LLM responses achieve targeted behavioral shifts in self-stigma support yet experts prefer generic empathy baselines.

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

The paper establishes that self-stigma among people who use drugs varies across four personas recovered from Reddit posts through latent profile analysis. Sequential classifiers identify these personas from short posting histories more accurately than batch or few-shot LLM methods. When clinical experts evaluated LLM replies across three models, the persona-conditioned versions produced the intended shifts in behavioral markers, but the same raters gave higher holistic scores to unconditioned generic empathy responses. This reveals a direct tension between responses optimized for specific clinical targets and those judged most empathetic overall. Readers care because conversational support tools for stigma must decide whether to prioritize measurable behavioral alignment or broad preference ratings.

Core claim

Persona-matched responses successfully achieved targeted behavioral shifts, yet raters holistically preferred the generic empathy of the persona-neutral baseline. The four-persona typology recovered from indicator-level features on 1,174 Reddit users enables this conditioning, and the misalignment indicates that holistic empathy judgments and clinically-aligned response design can pull in opposite directions.

What carries the argument

Four-persona typology from latent profile analysis on self-stigma indicator features, used both for sequential classification from limited histories and for conditioning LLM response generation.

If this is right

  • Sequential Bayesian and recurrent neural classifiers recover the four personas at macro-F1 of 0.74 from only 30 posts.
  • Persona-conditioned LLM outputs meet targeted behavioral-shift criteria set by clinical experts.
  • Holistic preference ratings favor persona-neutral generic empathy across eight raters and three LLMs.
  • Future evaluation rubrics for LLM stigma support must separate clinical alignment from overall empathy judgments.

Where Pith is reading between the lines

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

  • Systems could present users with a choice between tailored and generic reply styles rather than committing to one.
  • The same persona-detection pipeline might apply to other forms of internalized stigma such as mental-health or weight-related self-stigma.
  • Longitudinal deployment data would test whether preference ratings predict continued use of the support tool.

Load-bearing premise

The four-persona typology produced by latent profile analysis on Reddit self-stigma posts captures stable, clinically meaningful categories that can be detected from limited histories and used to design superior support responses.

What would settle it

A controlled trial in which actual people who use drugs interact with both persona-matched and generic responses and show no measurable difference in engagement, stigma reduction, or preference ratings between the two.

Figures

Figures reproduced from arXiv: 2606.23387 by Layla Bouzoubaa, Rezvaneh Rezapour.

Figure 1
Figure 1. Figure 1: Overview of our three-phase approach. Phases 2–3 (outer box) constitute the proposed support approach; [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Four self-stigma personas identified via LPA. Each card shows the top distinguishing [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Persona recovery (macro-F1) vs. posting history length. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Expected Calibration Error by model across [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
read the original abstract

Self-stigma predicts treatment avoidance and disengagement among people who use drugs (PWUD), yet conversational systems aiming to provide support typically treat self-stigma expression as a uniform signal. We present a three-phase, proof-of-concept study of a persona-aware approach to LLM support. Latent Profile Analysis (LPA) on indicator-level features from 1,174 self-stigma expressors on Reddit yields a four-persona typology validated against held-out behavioral and linguistic features. Sequential Bayesian and recurrent neural classifiers recover these personas from limited posting histories, substantially outperforming batch and few-shot LLM baselines (macro-F1 = 0.74 at 30 posts). Evaluation by eight clinical experts across three contemporary LLMs revealed a misalignment: persona-matched responses successfully achieved targeted behavioral shifts, yet raters holistically preferred the generic empathy of the persona-neutral baseline. Our findings suggest that holistic empathy judgments and clinically-aligned response design can pull in opposite directions, and that evaluating LLM-based stigma support requires rubrics capable of decomposing the two.

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.

Referee Report

3 major / 1 minor

Summary. The paper presents a three-phase proof-of-concept study claiming that Latent Profile Analysis (LPA) on indicator features from 1,174 Reddit self-stigma posts yields a stable four-persona typology of PWUD self-stigma. Sequential classifiers recover these personas from limited histories (macro-F1 0.74 at 30 posts, outperforming LLM baselines). Expert evaluation of LLM responses across three models shows persona-matched outputs achieve targeted behavioral shifts, yet raters prefer the holistic empathy of persona-neutral baselines, implying that clinically-aligned design and overall preference judgments can conflict and that standard evaluation rubrics are insufficient.

Significance. If the typology is shown to be stable and externally valid, the work usefully demonstrates a concrete misalignment between targeted behavioral outcomes and holistic preference in LLM support systems for stigmatized groups. The multi-phase design (LPA + classification + expert rating) and reported numbers (1,174 users, 8 experts) provide a replicable template for persona-conditioned conversational agents. Credit is due for the explicit comparison of persona-matched vs. generic conditions and for surfacing the tension between rubrics.

major comments (3)
  1. [LPA methods] LPA methods section: No details are provided on model selection criteria (BIC, AIC, entropy, or bootstrap likelihood ratio tests) or sensitivity analyses for the choice of exactly four profiles; without these, it is impossible to assess whether the typology is robust or an artifact of the indicator features, which directly undermines the downstream classifier training and the interpretation of the expert misalignment result.
  2. [Expert evaluation] Expert evaluation section: Inter-rater reliability (e.g., Fleiss' kappa or ICC) is not reported for the eight clinical experts' judgments of behavioral shifts versus holistic preference; this is load-bearing because the central claim rests on the existence of a reliable misalignment between the two rating dimensions.
  3. [Validation and results] Validation and results sections: Personas are validated solely against held-out linguistic/behavioral features from the same Reddit corpus and recovered via classifiers trained on LPA-derived labels; no external anchoring to clinical instruments, treatment outcomes, or independent expert clinical profiles is described, leaving open the possibility that the four-persona structure lacks clinical meaningfulness and that both the targeted-shift and misalignment findings are therefore ungrounded.
minor comments (1)
  1. [Abstract and methods] The abstract and methods should explicitly name the three contemporary LLMs used in the expert evaluation for reproducibility.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We appreciate the referee's constructive feedback on our manuscript. Below we provide point-by-point responses to the major comments and indicate the revisions we plan to make.

read point-by-point responses

  1. Referee: [LPA methods] LPA methods section: No details are provided on model selection criteria (BIC, AIC, entropy, or bootstrap likelihood ratio tests) or sensitivity analyses for the choice of exactly four profiles; without these, it is impossible to assess whether the typology is robust or an artifact of the indicator features, which directly undermines the downstream classifier training and the interpretation of the expert misalignment result.

    Authors: We agree with this assessment. The original submission omitted these details due to space constraints in the methods section. In the revised manuscript, we will expand the LPA methods to include model selection criteria (BIC, AIC, entropy, BLRT) and sensitivity analyses demonstrating the robustness of the four-profile solution. revision: yes

  2. Referee: [Expert evaluation] Expert evaluation section: Inter-rater reliability (e.g., Fleiss' kappa or ICC) is not reported for the eight clinical experts' judgments of behavioral shifts versus holistic preference; this is load-bearing because the central claim rests on the existence of a reliable misalignment between the two rating dimensions.

    Authors: We concur that reporting inter-rater reliability is essential. We will compute and include Fleiss' kappa (or appropriate ICC) for the expert ratings in the revised manuscript to substantiate the reliability of the observed misalignment. revision: yes

  3. Referee: [Validation and results] Validation and results sections: Personas are validated solely against held-out linguistic/behavioral features from the same Reddit corpus and recovered via classifiers trained on LPA-derived labels; no external anchoring to clinical instruments, treatment outcomes, or independent expert clinical profiles is described, leaving open the possibility that the four-persona structure lacks clinical meaningfulness and that both the targeted-shift and misalignment findings are therefore ungrounded.

    Authors: This is a valid concern for a study aiming at clinical relevance. Our work is framed as a proof-of-concept using publicly available Reddit data, with internal validation against held-out features. We cannot provide external clinical anchoring without new data collection involving clinical populations and instruments, which is outside the scope of the current study. In revision, we will strengthen the limitations section to discuss this gap and its implications for generalizability, while emphasizing the value of the digital trace approach for initial typology development. revision: partial

standing simulated objections not resolved
  • External clinical validation of the personas (new data collection required beyond current Reddit corpus)

Circularity Check

0 steps flagged

No significant circularity; derivation chain is self-contained

full rationale

The paper derives a four-persona typology via LPA on indicator features from 1,174 Reddit posts, validates it against held-out behavioral/linguistic features from the same corpus, trains classifiers to recover the LPA labels (macro-F1 reported on held-out data), and evaluates LLM responses via independent clinical expert ratings. No equations or steps reduce any claimed outcome (targeted behavioral shifts or holistic preference misalignment) to a fitted parameter of the same input data by construction. No load-bearing self-citations, uniqueness theorems, or ansatzes imported from prior author work are described. The central claim rests on external expert judgments rather than internal re-derivation of the personas themselves.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 1 invented entities

The typology rests on the assumption that latent profiles exist in the chosen indicator features; the number of profiles is a fitted parameter; personas are invented constructs without external validation mentioned.

free parameters (1)
  • number of latent profiles = 4
    Chosen via LPA on self-stigma indicator features from 1,174 users
axioms (2)
  • domain assumption Self-stigma expressions on Reddit form meaningful latent profiles detectable from linguistic and behavioral indicators
    Invoked by the use of LPA in phase 1
  • domain assumption Expert holistic ratings of LLM responses are a valid proxy for support quality
    Used to conclude misalignment with targeted behavioral shifts
invented entities (1)
  • four self-stigma personas no independent evidence
    purpose: To enable persona-conditioned LLM support
    Derived from LPA clustering without mentioned external clinical validation

pith-pipeline@v0.9.1-grok · 5719 in / 1374 out tokens · 43680 ms · 2026-06-26T08:04:31.380552+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

293 extracted references · 162 canonical work pages · 2 internal anchors

  1. [1]

    and Takano, Keisuke and Yu, Placida Hoi Man and Wong, Patrina Hei Tung and Barry, Tom J

    Adelina, Nadia and Chan, Christian S. and Takano, Keisuke and Yu, Placida Hoi Man and Wong, Patrina Hei Tung and Barry, Tom J. , date =. The. 2023 , journaltitle =. doi:10.1089/cyber.2023.0144 , url =

    work page doi:10.1089/cyber.2023.0144 2023

  2. [2]

    2019 , journaltitle =

    A Text Classification Framework for Simple and Effective Early Depression Detection over Social Media Streams , author =. 2019 , journaltitle =. doi:10.1016/j.eswa.2019.05.023 , url =

    work page doi:10.1016/j.eswa.2019.05.023 2019

  3. [3]

    2018 , doi =

    Cohan, Arman and Desmet, Bart and Yates, Andrew and Soldaini, Luca and MacAvaney, Sean and Goharian, Nazli , date =. 2018 , doi =

    2018

  4. [5]

    and Sedig, Kamran and Lizotte, Daniel J

    Davis, Brent D. and Sedig, Kamran and Lizotte, Daniel J. , date =. Archetype-. 2019 , journaltitle =. doi:10.3390/bdcc3030044 , url =

    work page doi:10.3390/bdcc3030044 2019

  5. [7]

    Heo, Ruth and Depp, Colin , date =. The. 2026 , journaltitle =. doi:10.3928/00485713-20260416-02 , url =

    work page doi:10.3928/00485713-20260416-02 2026

  6. [8]

    Understanding

    Kim, Seoyun and Cha, Junyeop and Kim, Dongjae and Park, Eunil , date =. Understanding. 2023 , journaltitle =. doi:10.2196/49074 , url =

    work page doi:10.2196/49074 2023

  7. [10]

    and Wenzel, Michael and Haines, Emily and Stevens, Jesse and Fighera, Daniel and Williams, Patrick and Arthurson, Samuel and Osborne, Danny and Skitka, Linda J

    Lizzio‐Wilson, Morgana and Thomas, Emma F. and Wenzel, Michael and Haines, Emily and Stevens, Jesse and Fighera, Daniel and Williams, Patrick and Arthurson, Samuel and Osborne, Danny and Skitka, Linda J. , date =. What Could Be?. 2025 , journaltitle =. doi:10.1111/bjso.12853 , url =. 39898497 , eprinttype =

    work page doi:10.1111/bjso.12853 2025

  8. [12]

    Low, Daniel M and Rumker, Laurie and Talkar, Tanya and Torous, John and Cecchi, Guillermo and Ghosh, Satrajit S , date =. Natural. 2020 , journaltitle =. doi:10.2196/22635 , url =

    work page doi:10.2196/22635 2020

  9. [13]

    , author =

    Ten Frequently Asked Questions about Latent Class Analysis. , author =. 2018 , journaltitle =. doi:10.1037/tps0000176 , url =

    work page doi:10.1037/tps0000176 2018

  10. [14]

    Harnessing the

    Qian, Yushan and Zhang, Weinan and Liu, Ting , date =. Harnessing the. Findings of the. 2023 , pages =. doi:10.18653/v1/2023.findings-emnlp.433 , url =

    work page doi:10.18653/v1/2023.findings-emnlp.433 2023

  11. [15]

    Understanding Mental Health Discourse on

    Sánchez Rodríguez, Irene and Bianchi, John and Pinelli, Fabio and Panizza, Folco and Ricciardi, Emiliano and Pietrini, Pietro and Petrocchi, Marinella , date =. Understanding Mental Health Discourse on. 2026 , journaltitle =. doi:10.1038/s41598-026-35918-3 , url =

    work page doi:10.1038/s41598-026-35918-3 2026

  12. [16]

    A Computational Approach to Understanding Empathy Expressed in Text-Based Mental Health Support , booktitle =

    Sharma, Ashish and Miner, Adam and Atkins, David and Althoff, Tim , editor =. A Computational Approach to Understanding Empathy Expressed in Text-Based Mental Health Support , booktitle =. 2020 , pages =. doi:10.18653/v1/2020.emnlp-main.425 , url =

    work page doi:10.18653/v1/2020.emnlp-main.425 2020

  13. [17]

    Cognitive

    Sharma, Ashish and Rushton, Kevin and Lin, Inna and Wadden, David and Lucas, Khendra and Miner, Adam and Nguyen, Theresa and Althoff, Tim , date =. Cognitive. Proceedings of the 61st. 2023 , pages =. doi:10.18653/v1/2023.acl-long.555 , url =

    work page doi:10.18653/v1/2023.acl-long.555 2023

  14. [19]

    Wang, Hongru and Wang, Rui and Mi, Fei and Deng, Yang and Wang, Zezhong and Liang, Bin and Xu, Ruifeng and Wong, Kam-Fai , date =. Cue-. Findings of the. 2023 , pages =. doi:10.18653/v1/2023.findings-emnlp.806 , url =

    work page doi:10.18653/v1/2023.findings-emnlp.806 2023

  15. [20]

    Identifying

    Wang, Lexie Chenyue and Pike, Kenneth C and Conway, Mike and Chen, Annie T , date =. Identifying. 2025 , journaltitle =. doi:10.2196/68695 , url =

    work page doi:10.2196/68695 2025

  16. [21]

    Proceedings of the 34th

    Wang, Xi and Perez, Anxo and Parapar, Javier and Crestani, Fabio , date =. Proceedings of the 34th. 2025 , pages =. doi:10.1145/3746252.3761617 , url =

    work page doi:10.1145/3746252.3761617 2025

  17. [22]

    , editor =

    Wei, Yangbo and Huang, Zhen and Zhao, Fangzhou and Feng, Qi and Xing, Wei W. , editor =. Findings of the. 2025 , pages =. doi:10.18653/v1/2025.findings-acl.435 , url =

    work page doi:10.18653/v1/2025.findings-acl.435 2025

  18. [23]

    Wu, Shenghan and Zhu, Yimo and Hsu, Wynne and Lee, Mong-Li and Deng, Yang , editor =. From. Proceedings of the 2025. 2025 , pages =. doi:10.18653/v1/2025.emnlp-main.277 , url =

    work page doi:10.18653/v1/2025.emnlp-main.277 2025

  19. [24]

    Zhong, Peixiang and Zhang, Chen and Wang, Hao and Liu, Yong and Miao, Chunyan , editor =. Towards. Proceedings of the 2020. 2020 , pages =. doi:10.18653/v1/2020.emnlp-main.531 , url =

    work page doi:10.18653/v1/2020.emnlp-main.531 2020

  20. [25]

    Detecting

    Wolohan,. Detecting. Proceedings of the

  21. [26]

    The dawn of quantum natural language processing,

    Lian, Ruixue and Huang, Che-Wei and Tang, Yuqing and Gu, Qilong and Ma, Chengyuan and Guo, Chenlei , date =. Incremental. doi:10.1109/ICASSP43922.2022.9747689 , url =

    work page doi:10.1109/icassp43922.2022.9747689 2022

  22. [27]

    Harrigian, Keith and Aguirre, Carlos and Dredze, Mark , date =. Do. Findings of the. doi:10.18653/v1/2020.findings-emnlp.337 , url =

    work page doi:10.18653/v1/2020.findings-emnlp.337 2020

  23. [28]

    Harrigian, Keith and Aguirre, Carlos and Dredze, Mark , date =. On the. Proceedings of the. doi:10.18653/v1/2021.clpsych-1.2 , url =

    work page doi:10.18653/v1/2021.clpsych-1.2 2021

  24. [29]

    SMHD: A Large-Scale Resource for Exploring Online Language Usage for Multiple Mental Health Conditions

    Cohan, Arman and Desmet, Bart and Yates, Andrew and Soldaini, Luca and MacAvaney, Sean and Goharian, Nazli , date =. doi:10.48550/ARXIV.1806.05258 , url =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1806.05258

  25. [30]

    Dalal, Sumit and Jain, Sarika and Dave, Mayank , date =. An. doi:10.2174/1872212117666220812110956 , url =

    work page doi:10.2174/1872212117666220812110956

  26. [31]

    and Crestani, Fabio and Parapar, Javier , editor =

    Losada, David E. and Crestani, Fabio and Parapar, Javier , editor =. Experimental. doi:10.1007/978-3-319-65813-1_30 , url =

    work page doi:10.1007/978-3-319-65813-1_30

  27. [32]

    Ferro, Nicola and Peters, Carol , editor =. From. Information. doi:10.1007/978-3-030-22948-1_1 , url =

    work page doi:10.1007/978-3-030-22948-1_1

  28. [33]

    and Errecalde, Marcelo and Montes-y-Gómez, Manuel , date =

    Burdisso, Sergio G. and Errecalde, Marcelo and Montes-y-Gómez, Manuel , date =. T-. doi:10.48550/ARXIV.1911.06147 , url =

    work page doi:10.48550/arxiv.1911.06147 1911

  29. [34]

    Representation

    Ji, Yangfeng and Eisenstein, Jacob , date =. Representation. Proceedings of the 52nd. doi:10.3115/v1/P14-1002 , url =

    work page doi:10.3115/v1/p14-1002

  30. [35]

    Gkotsis, George and Oellrich, Anika and Velupillai, Sumithra and Liakata, Maria and Hubbard, Tim J. P. and Dobson, Richard J. B. and Dutta, Rina , date =. Characterisation of Mental Health Conditions in Social Media Using. doi:10.1038/srep45141 , url =

    work page doi:10.1038/srep45141

  31. [36]

    Analysing Relevance of

    Agarwal, Navneet and Dias, Gaël and Dollfus, Sonia , date =. Analysing Relevance of. Proceedings of the 9th. doi:10.18653/v1/2024.clpsych-1.9 , url =

    work page doi:10.18653/v1/2024.clpsych-1.9 2024

  32. [37]

    Liu, Xingyun and Liu, Xiaoqian , date =. Online. doi:10.3390/healthcare9070847 , url =

    work page doi:10.3390/healthcare9070847

  33. [38]

    Predicting

    De Choudhury, Munmun and Gamon, Michael and Counts, Scott and Horvitz, Eric , date =. Predicting. doi:10.1609/icwsm.v7i1.14432 , url =

    work page doi:10.1609/icwsm.v7i1.14432

  34. [39]

    doi:10.48550/ARXIV.2110.15621 , url =

    Ji, Shaoxiong and Zhang, Tianlin and Ansari, Luna and Fu, Jie and Tiwari, Prayag and Cambria, Erik , date =. doi:10.48550/ARXIV.2110.15621 , url =

    work page doi:10.48550/arxiv.2110.15621

  35. [40]

    Proceedings of the

    Yang, Kailai and Zhang, Tianlin and Kuang, Ziyan and Xie, Qianqian and Huang, Jimin and Ananiadou, Sophia , date =. Proceedings of the. doi:10.1145/3589334.3648137 , url =

    work page doi:10.1145/3589334.3648137

  36. [41]

    doi:10.48550/ARXIV.2511.04698 , url =

    Islam, K M Sajjadul and Fields, John and Madiraju, Praveen , date =. doi:10.48550/ARXIV.2511.04698 , url =

    work page doi:10.48550/arxiv.2511.04698

  37. [42]

    Hasan, Khalid and Saquer, Jamil and Zhang, Yifan , date =. Mental. doi:10.48550/ARXIV.2509.16542 , url =

    work page doi:10.48550/arxiv.2509.16542

  38. [43]

    Exploring

    Zanwar, Sourabh and Wiechmann, Daniel and Qiao, Yu and Kerz, Elma , date =. Exploring. Proceedings of the 13th. doi:10.18653/v1/2022.louhi-1.21 , url =

    work page doi:10.18653/v1/2022.louhi-1.21 2022

  39. [44]

    and Wang, Dakuo , date =

    Xu, Xuhai and Yao, Bingsheng and Dong, Yuanzhe and Gabriel, Saadia and Yu, Hong and Hendler, James and Ghassemi, Marzyeh and Dey, Anind K. and Wang, Dakuo , date =. Mental-. doi:10.1145/3643540 , url =

    work page doi:10.1145/3643540

  40. [45]

    Ratcliff, Roger and McKoon, Gail , date =. The. doi:10.1162/neco.2008.12-06-420 , url =. 18085991 , eprinttype =

    work page doi:10.1162/neco.2008.12-06-420 2008

  41. [46]

    Sequential

    Bapna, Ankur and Tür, Gokhan and Hakkani-Tür, Dilek and Heck, Larry , editor =. Sequential. Proceedings of the 18th. doi:10.18653/v1/W17-5514 , url =

    work page doi:10.18653/v1/w17-5514

  42. [47]

    Basit, Mohammad and Alam, Bashir and Fatima, Zubaida and Shaikh, Salman , editor =. Natural. Proceedings of the 2023. doi:10.18653/v1/2023.emnlp-main.471 , url =

    work page doi:10.18653/v1/2023.emnlp-main.471 2023

  43. [48]

    Cho, Itsugun and Wang, Dongyang and Takahashi, Ryota and Saito, Hiroaki , editor =. A. Proceedings of the 29th. 2022 , date =

    2022

  44. [49]

    Passive-

    Derczynski, Leon and Bontcheva, Kalina , editor =. Passive-. Proceedings of the 14th. doi:10.3115/v1/E14-4014 , url =

    work page doi:10.3115/v1/e14-4014

  45. [50]

    Hierarchical Neural Model with Attention Mechanisms for the Classification of Social Media Text Related to Mental Health , booktitle =

    Ive, Julia and Gkotsis, George and Dutta, Rina and Stewart, Robert and Velupillai, Sumithra , editor =. Hierarchical Neural Model with Attention Mechanisms for the Classification of Social Media Text Related to Mental Health , booktitle =. doi:10.18653/v1/W18-0607 , url =

    work page doi:10.18653/v1/w18-0607

  46. [51]

    Lee, Honghee and Ko, Youngjoong , date =. Dynamic. doi:10.1016/j.csl.2025.101896 , url =

    work page doi:10.1016/j.csl.2025.101896 2025

  47. [52]

    End-to-End

    Ma, Xuezhe and Hovy, Eduard , editor =. End-to-End. Proceedings of the 54th. doi:10.18653/v1/P16-1101 , url =

    work page doi:10.18653/v1/p16-1101

  48. [53]

    Olabiyi, Oluwatobi and Khazane, Anish and Salimov, Alan and Mueller, Erik , editor =. An. Proceedings of the. doi:10.18653/v1/W19-2301 , url =

    work page doi:10.18653/v1/w19-2301

  49. [54]

    Peng, Xingyu and Wu, Junran and Liu, Ruomei and Xu, Ke , editor =. Rumor. Proceedings of the 31st

  50. [55]

    Tamire, Maunika and Anumasa, Srinivas and Srijith, P. K. , date =. Bi-. doi:10.48550/arXiv.2112.12809 , url =. 2112.12809 , eprinttype =

    work page doi:10.48550/arxiv.2112.12809

  51. [56]

    , author=

    Properties of the hubert-arable adjusted rand index. , author=. Psychological methods , volume=. 2004 , publisher=

    2004

  52. [57]

    Translational Issues in Psychological Science , year=

    Ten Frequently Asked Questions About Latent Class Analysis , author=. Translational Issues in Psychological Science , year=

  53. [58]

    , author=

    A new readability yardstick. , author=. The Journal of applied psychology , year=

  54. [59]

    2016 , publisher=

    Latent class analysis and latent profile analysis , author=. 2016 , publisher=

    2016

  55. [60]

    Structural equation modeling: A multidisciplinary Journal , volume=

    Auxiliary variables in mixture modeling: Three-step approaches using M plus , author=. Structural equation modeling: A multidisciplinary Journal , volume=. 2014 , publisher=

    2014

  56. [61]

    and Watson, Amy C

    Corrigan, Patrick W. and Watson, Amy C. and Barr, Leah , title =. Journal of Social and Clinical Psychology , volume =. 2006 , doi =

    2006

  57. [62]

    and Grajales, Monica , title =

    Ritsher, Jennifer Boyd and Otilingam, Poorni G. and Grajales, Monica , title =. Psychiatry Research , volume =. 2003 , doi =

    2003

  58. [63]

    and Rao, Deepa , title =

    Corrigan, Patrick W. and Rao, Deepa , title =. Canadian Journal of Psychiatry , volume =. 2012 , doi =

    2012

  59. [64]

    Brendan and Raftery, Adrian E

    Scrucca, Luca and Fop, Michael and Murphy, T. Brendan and Raftery, Adrian E. , title =. The R Journal , volume =. 2016 , doi =

    2016

  60. [65]

    and Asparouhov, Tihomir and Muth\'

    Nylund, Karen L. and Asparouhov, Tihomir and Muth\'. Deciding on the Number of Classes in Latent Class Analysis and Growth Mixture Modeling:. Structural Equation Modeling , volume =. 2007 , doi =

    2007

  61. [66]

    and Ashokkumar, Ashwini and Seraj, Sarah and Pennebaker, James W

    Boyd, Ryan L. and Ashokkumar, Ashwini and Seraj, Sarah and Pennebaker, James W. , title =. 2022 , note =

    2022

  62. [67]

    Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work and Social Computing (CSCW) , pages =

    MacLean, Diana and Gupta, Sonal and Lembke, Anna and Manning, Christopher and Heer, Jeffrey , title =. Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work and Social Computing (CSCW) , pages =. 2015 , doi =

    2015

  63. [68]

    Proceedings of the 8th International AAAI Conference on Web and Social Media (ICWSM) , pages =

    De Choudhury, Munmun and De, Sushovan , title =. Proceedings of the 8th International AAAI Conference on Web and Social Media (ICWSM) , pages =

  64. [69]

    Journal of the Royal Statistical Society: Series B , volume =

    Benjamini, Yoav and Hochberg, Yosef , title =. Journal of the Royal Statistical Society: Series B , volume =. 1995 , doi =

    1995

  65. [70]

    Frontiers in Psychology , volume =

    Lakens, Daniel , title =. Frontiers in Psychology , volume =. 2013 , doi =

    2013

  66. [71]

    and Phelan, Jo C

    Link, Bruce G. and Phelan, Jo C. , title =. Annual Review of Sociology , volume =. 2001 , doi =

    2001

  67. [72]

    Assessing the

    Badawi, Abeer and Laskar, Md Tahmid Rahman and Rahimi, Elahe and Grach, Sheri and Bertrand, Lindsay and Danok, Lames and Rudzicz, Frank and Huang, Jimmy and Dolatabadi, Elham , month = jan, year =. Assessing the. doi:10.48550/arXiv.2601.18630 , abstract =

    work page doi:10.48550/arxiv.2601.18630

  68. [73]

    Badawi, Abeer and Rahimi, Elahe and Laskar, Md Tahmid Rahman and Grach, Sheri and Bertrand, Lindsay and Danok, Lames and Dhanesh, Prathiba and Huang, Jimmy and Rudzicz, Frank and Dolatabadi, Elham , editor =. When. Proceedings of the 19th. 2026 , pages =. doi:10.18653/v1/2026.eacl-long.180 , abstract =

    work page doi:10.18653/v1/2026.eacl-long.180 2026

  69. [74]

    and Groh, Matthew , month = feb, year =

    Kumar, Aakriti and Poungpeth, Nalin and Yang, Diyi and Farrell, Erina and Lambert, Bruce L. and Groh, Matthew , month = feb, year =. When large language models are reliable for judging empathic communication , volume =. Nature Machine Intelligence , publisher =. doi:10.1038/s42256-025-01169-6 , abstract =

    work page doi:10.1038/s42256-025-01169-6

  70. [75]

    Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , author =

    How. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , author =. 2025 , pages =. doi:10.1609/aies.v8i2.36632 , abstract =

    work page doi:10.1609/aies.v8i2.36632 2025

  71. [76]

    The Annals of Applied Statistics , author =

    Gelman, Andrew and Jakulin, Aleks and Pittau, Maria Grazia and Su, Yu-Sung , year=. A weakly informative default prior distribution for logistic and other regression models , volume=. The Annals of Applied Statistics , publisher=. doi:10.1214/08-aoas191 , number=

    work page doi:10.1214/08-aoas191

  72. [77]

    2026 , month = mar, url =

    Introducing. 2026 , month = mar, url =

    2026

  73. [78]

    2014 , publisher=

    Handbook of inter-rater reliability: The definitive guide to measuring the extent of agreement among raters , author=. 2014 , publisher=

    2014

  74. [79]

    The problems of two paradoxes , author=

    High agreement but low kappa: I. The problems of two paradoxes , author=. Journal of clinical epidemiology , volume=. 1990 , publisher=

    1990

  75. [80]

    and Demiris, George and Huh-Yoo, Jina and Rezapour, Rezvaneh

    Aghakhani, Elham and Wang, Lu and Washington, Karla T. and Demiris, George and Huh-Yoo, Jina and Rezapour, Rezvaneh. From Conversation to Automation: Leveraging LLM s for Problem-Solving Therapy Analysis. Findings of the Association for Computational Linguistics: ACL 2025. 2025. doi:10.18653/v1/2025.findings-acl.1292

    work page doi:10.18653/v1/2025.findings-acl.1292 2025

  76. [81]

    arXiv preprint arXiv:2601.20747 , year=

    Like a Therapist, But Not: Reddit Narratives of AI in Mental Health Contexts , author=. arXiv preprint arXiv:2601.20747 , year=

    arXiv

  77. [82]

    and Haber, Nick , month = jun, year =

    Moore, Jared and Grabb, Declan and Agnew, William and Klyman, Kevin and Chancellor, Stevie and Ong, Desmond C. and Haber, Nick , title =. Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency , pages =. 2025 , isbn =. doi:10.1145/3715275.3732039 , abstract =

    work page doi:10.1145/3715275.3732039 2025

  78. [83]

    and Rezapour, R

    Roshanaei, M. and Rezapour, R. and Seif El-Nasr, M. , title =. AI & Society , volume =. 2026 , doi =

    2026

  79. [84]

    The Innovation , publisher=

    A survey on llm-as-a-judge , author=. The Innovation , publisher=

  80. [85]

    , date =

    Yin, Yidan and Jia, Nan and Wakslak, Cheryl J. , date =. doi:10.1073/pnas.2319112121 , url =

    work page doi:10.1073/pnas.2319112121

Showing first 80 references.