Pith. sign in

REVIEW 2 major objections 4 minor 54 references

ADHD and ASD Twitter users emphasise different DSM-5 depressive symptoms at population scale, and the pattern holds across filtering thresholds.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

T0 review · grok-4.5

2026-07-11 04:43 UTC pith:JCMETP23

load-bearing objection Careful exploratory digital-phenotyping paper with a real multi-threshold robustness design and open artifacts; modest language-level ADHD/ASD depressive-symptom leanings that hold as scoped, not as clinical phenomenology. the 2 major comments →

arxiv 2607.05626 v1 pith:JCMETP23 submitted 2026-07-06 cs.CL

Population-Level Profiling of DSM-5 Depressive Symptoms Among Self-Reported ADHD and ASD Users on Twitter: An Exploratory Study Using Advanced NLP and Statistical Analysis

classification cs.CL
keywords depressionDSM-5ADHDautism spectrum disorderdigital phenotypingMentalRoBERTasocial medianatural language processing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper asks whether people who publicly self-report ADHD versus ASD on Twitter talk about depression differently when their posts are scored against the nine DSM-5 depressive symptoms. Using more than a million tweets from 792 users, the authors first screen for depressive relevance with a zero-shot model, then score each tweet for the nine symptoms with a domain-adapted language model fine-tuned on expert-annotated sentences. After centering each user’s profile so only relative symptom emphasis remains, an L1 logistic model modestly but stably separates the two groups (ROC-AUC 0.645–0.653). Cognitive complaints, sleep disturbance, appetite change, and fatigue lean toward ADHD; suicidal ideation and anhedonia lean toward ASD, with the same directional pattern surviving five different pre-filter thresholds and high bootstrap selection. Symptom co-occurrence structure is largely shared across groups; no pairwise correlation meets the authors’ strict test for a disorder-specific difference. The result is framed as a reproducible population-level language signal, not as clinical proof of different depressive phenomenology in individuals.

Core claim

At population level, self-reported ADHD and ASD Twitter users show consistent differences in relative emphasis among the nine DSM-5 depressive symptoms: cognitive issues, sleep issues, appetite change, and fatigue lean ADHD while suicidal ideation and anhedonia lean ASD. These directional effects remain stable across five depressive-content pre-filter thresholds with bootstrap selection frequency at least 0.90, while overall discrimination stays modest (cross-validated ROC-AUC 0.645–0.653) and the pairwise co-occurrence structure is largely shared.

What carries the argument

A two-stage pipeline (zero-shot NLI depression-relevance pre-filter plus MentalRoBERTa multi-label DSM-5 symptom classifier) that produces per-user mean-centered nine-symptom profiles, followed by L1-penalised logistic regression and a graded robustness scheme requiring identical coefficient sign and bootstrap selection ≥0.90 at every threshold.

Load-bearing premise

That public self-reports of an ADHD or ASD diagnosis on Twitter can be treated as valid group labels for the two disorders.

What would settle it

A replication on a large sample of users whose ADHD or ASD diagnoses are independently clinically verified, using the same centered symptom profiles and threshold-sweep protocol, that either reverses the six Level-1 coefficient signs or drives bootstrap selection below 0.90.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper analyses 1,282,437 tweets from 792 self-reported ADHD (n=622) and ASD (n=170) Twitter users. After zero-shot NLI depression-relevance pre-filtering and multi-label classification of nine DSM-5 depressive symptoms with MentalRoBERTa fine-tuned on ReDSM5 (held-out macro-F1 0.901), user-level mean-centered symptom profiles are formed. L1-penalised logistic regression with nested CV and 1,000-resample bootstrap stability, repeated across five pre-filter thresholds (0.45–0.65), yields modest but stable discrimination (CV ROC-AUC 0.645–0.653). Six Level-1 symptoms show consistent direction and selection ≥0.90 at every threshold: cognitive issues, sleep, appetite, and fatigue lean ADHD; suicidal ideation and anhedonia lean ASD. Pairwise co-occurrence is largely shared (17/36 pairs); no pair meets the pre-specified disorder-specific criterion. The authors frame the work as exploratory population-level language reproducibility, not clinical validity.

Significance. If the scoped claim holds, the work supplies a carefully controlled, multi-threshold digital-phenotyping characterisation of how depressive language is differentially emphasised in two large neurodevelopmental communities online. Strengths that raise the contribution above a typical social-media NLP study include: (i) an external expert-annotated training corpus (ReDSM5) rather than circular self-labels for symptoms, (ii) a pre-specified graded robustness scheme combining bootstrap selection frequency with cross-threshold sign consistency, (iii) user-level centering that isolates relative symptom emphasis, and (iv) public code, checkpoints, and calibrated thresholds. The modest AUC is expected and honestly reported; the value lies in reproducible directional patterns rather than predictive power. External validity remains limited by self-report diagnosis labels and Reddit-to-Twitter domain transfer, both already flagged in Limitations.

major comments (2)
  1. [§2.1.1 Data Sources; Limitations] §2.1.1 and Limitations: group membership rests entirely on unverified public diagnosis-disclosure tweets. The modest Level-1 coefficient directions and AUC 0.645–0.653 are therefore conditional on label fidelity. A sensitivity analysis that either (a) restricts to users with multiple independent disclosure statements or (b) reports the fraction of dual-labeled / ambiguous accounts would materially strengthen the claim that the signal maps to ADHD vs ASD rather than to disclosure style or comorbidity confounds.
  2. [§3.2 Classifier Performance; Table 2] §3.2 / Table 2: MentalRoBERTa is trained and evaluated only on Reddit sentences (ReDSM5) yet applied to tweets. Macro-F1 0.901 is strong on the held-out Reddit split, but no tweet-level human validation or cross-domain calibration set is provided. For the sparsest labels (psychomotor n=35, appetite n=48) and for the Level-1 appetite finding, even modest domain shift could reverse or attenuate coefficients. A small manually annotated Twitter subsample (or at least error analysis of high-activation tweets) is needed before the directional claims can be treated as measurement-robust.
minor comments (4)
  1. [Table 2; §3.3] Table 2 reports Appetite Change test F1 = 1.000 on n=48 support; the authors correctly caution, but the main text still lists it among Level-1 primary findings without repeating the support caveat in the same sentence as the coefficient claim.
  2. [Figure 2] Figure 2 caption and legend are dense; the distinction between solid markers (CI excludes zero) and selection-frequency labels could be clarified with a short key in the figure itself.
  3. [§3.1; Supplementary Table S7] Supplementary Table S7 is essential for understanding attrition; a one-sentence pointer in the main-text Corpus Overview would help readers locate the full pipeline counts.
  4. [Ethics Statement] The Ethics Statement and Data Availability are exemplary; consider adding a brief note on whether any automated content filters (e.g., platform suicide-prevention redaction) could systematically under-sample suicidal-ideation language.

Circularity Check

0 steps flagged

No significant circularity: external ReDSM5 training labels, independent self-report group labels, and post-hoc discrimination/co-occurrence on held-out Twitter profiles do not reduce by construction to their inputs.

full rationale

The derivation chain is self-contained and non-circular. Stage-2 MentalRoBERTa is fine-tuned and evaluated solely on the external expert-annotated ReDSM5 Reddit corpus (held-out macro-F1 0.901), then applied zero-shot to Twitter; group membership (ADHD vs ASD) is taken from an independent diagnosis-disclosure dataset and is never used as a training target for the symptom classifier. User-level mean-centering, L1-logistic coefficients, bootstrap selection frequencies, and Pearson co-occurrence matrices are computed after the fact on the resulting profiles; the modest CV ROC-AUC (0.645–0.653) and Level-1 leanings are therefore empirical outputs, not tautologies forced by the fitting procedure or by any self-citation. Threshold sweeps and pre-specified robustness criteria are sensitivity checks, not definitional identities. No uniqueness theorem, ansatz, or load-bearing self-citation appears. Domain-transfer and self-report validity risks are external-validity issues already flagged in Limitations, not internal circularity.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 0 invented entities

The central claim rests on treating self-disclosed Twitter diagnoses as group labels, on transfer of a Reddit-trained multi-label symptom model to tweets, and on several analysis gates and thresholds chosen by the authors. No new physical entities are postulated; free parameters are pipeline cutoffs and regularisation rather than fitted scientific constants. Domain assumptions about DSM-5 symptom expressibility in short social posts and about NLI as a high-recall pre-filter are load-bearing but standard in this literature.

free parameters (5)
  • Stage-1 zero-shot NLI positive probability thresholds = 0.45, 0.50, 0.55, 0.60, 0.65
    Five cutoffs 0.45–0.65 control which tweets enter Stage 2; results are required to be sign-stable across them, but the range itself is author-chosen.
  • Per-label MentalRoBERTa decision thresholds = e.g. Suicidal 0.150; Depressed Mood 0.825
    Validation-set F1-maximising thresholds (0.150–0.825) convert continuous scores to binary gates and evaluation labels; they are fitted to ReDSM5 validation data.
  • Per-user minimum gated-tweet count = 30
    Users with fewer than 30 doubly-gated tweets are dropped; the number is a design choice justified by SE bounds, not estimated from a loss.
  • L1 logistic inverse regularisation C = CV-selected per threshold
    Chosen by 5-fold CV ROC-AUC over a 20-point log grid at each threshold; values vary (e.g. 1.62–206.91 in Table S1).
  • Stage-2 training hyperparameters (encoder LR, dropout, positive-weight boost, warmup) = LR 2e-5; dropout 0.4; boost 1.5; warmup 0.10; epoch 9
    Selected by 48-cell grid search on validation macro-F1 then fixed for full retrain.
axioms (6)
  • domain assumption Public self-report of a clinical ADHD or ASD diagnosis on Twitter is a usable proxy for group membership in population-level comparison.
    Invoked in Data Sources §2.1.1 and throughout group analyses; Limitations explicitly notes misdiagnosis and self-diagnosis risk.
  • domain assumption Sentence-level Reddit annotations of DSM-5 symptoms transfer sufficiently to Twitter for multi-label scoring after domain-adapted pretraining.
    Stated under Annotation Corpus and Limitations; underpins all user profiles.
  • ad hoc to paper User-level mean-centering isolates relative symptom emphasis independent of overall depressive-language intensity.
    Methods §2.3.1; this transformation defines the features fed to logistic regression and correlations.
  • domain assumption Zero-shot NLI against the hypothesis 'DSM-5 depressive symptom present' is a valid high-recall relevance screen.
    Stage 1 §2.2.1; justified by general NLI benchmarks rather than a labeled Twitter depression-relevance set.
  • domain assumption DSM-5 nine-symptom major depressive episode criteria are the right axes for population language profiling.
    Background and ReDSM5 labeling scheme; standard clinical taxonomy applied as multi-label targets.
  • ad hoc to paper Bootstrap selection frequency ≥0.90 at every threshold plus sign consistency defines 'robust' findings.
    Pre-specified graded robustness scheme §2.3.4; determines which symptoms are interpreted as Level 1.

pith-pipeline@v1.1.0-grok45 · 24993 in / 3801 out tokens · 36025 ms · 2026-07-11T04:43:25.043727+00:00 · methodology

0 comments
read the original abstract

Background: Depression frequently co-occurs with ADHD and autism spectrum disorder (ASD), but population-level differences in symptom expression between these groups remain underexplored. Objective: We examined whether social media users with ADHD and ASD differ in how they express DSM-5 depressive symptoms in their tweets, and whether differences persist across varying levels of depressive-content filtering. Methods: We analysed 1,282,437 tweets from 792 users (622 ADHD; 170 ASD) with self-reported diagnoses on Twitter. Tweets were pre-filtered for depressive relevance using zero-shot NLI, then classified into nine DSM-5 symptoms using MentalRoBERTa fine-tuned on ReDSM5. Profiles were mean-centered per user. We applied L1-penalised logistic regression with cross-validation to distinguish ADHD from ASD users, complemented by Pearson correlations for symptom co-occurrence, and tested robustness across five filtering thresholds using bootstrapping. Results: MentalRoBERTa achieved macro-F1 of 0.901 on a held-out set, outperforming the original ReDSM5 benchmark. ADHD vs ASD classification yielded stable but modest performance (cross-validated ROC-AUC 0.645-0.653). Cognitive issues, sleep issues, appetite change, and fatigue leaned toward ADHD, while suicidal ideation and anhedonia leaned toward ASD. A largely shared symptom co-occurrence structure emerged between groups; no pair met our criterion for a robust disorder-specific difference. Conclusions: Population-level differences in depression-related language between ADHD and ASD social media users were consistently observed across thresholds, reflecting reproducibility rather than clinical validity. Findings are exploratory and do not establish differing phenomenology at the individual level.

Figures

Figures reproduced from arXiv: 2607.05626 by David Nabergoj, Jure Dem\v{s}ar, Muhammad Rizwan.

Figure 1
Figure 1. Figure 1: Study workflow for detecting depressive symptom patterns in ADHD and ASD Twitter users. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Bootstrap-mean coefficient trajectory across depressive pre-filter thresholds, with bootstrap selection [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: ADHD versus ASD differential symptom co-occurrence. [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗

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

54 extracted references · 19 canonical work pages

  1. [1]

    Adult ADHD and comorbid de- pression: a consensus-derived diagnostic algorithm for ADHD

    McIntosh D, Kutcher S, Binder C, Levitt A, Fallu A, Rosenbluth M. Adult ADHD and comorbid de- pression: a consensus-derived diagnostic algorithm for ADHD. Neuropsychiatr Dis Treat 2009;5:137–

  2. [2]

    doi:https://doi.org/10.2147/ndt.s4720

  3. [3]

    Anxiety and depression in adults with autism spectrum disorder: a systematic review and meta-analysis

    Hollocks MJ, Lerh JW, Magiati I, Meiser-Stedman R, Brugha TS. Anxiety and depression in adults with autism spectrum disorder: a systematic review and meta-analysis. Psychol Med 2019;49(4):559–

  4. [4]

    doi:https://doi.org/10.1017/s0033291718002283

  5. [5]

    Does executive functioning (EF) predict depression in clinic- referred adults?: EF tests vs

    Knouse LE, Barkley RA, Murphy KR. Does executive functioning (EF) predict depression in clinic- referred adults?: EF tests vs. rating scales. J Affect Disord 2012;145(2):270–275. doi:https:// doi.org/10.1016/j.jad.2012.05.064

  6. [6]

    Having All of Your Internal Resources Exhausted Beyond Measure and Being 15 Left with No Clean-Up Crew

    Raymaker DM, Teo AR, Steckler NA, Lentz B, Scharer M, Delos Santos A, Kapp SK, Hunter M, Joyce A, Nicolaidis C. “Having All of Your Internal Resources Exhausted Beyond Measure and Being 15 Left with No Clean-Up Crew”: Defining autistic burnout. Autism Adulthood 2020;2(2):132–143. doi: https://doi.org/10.1089/aut.2019.0079

  7. [7]

    Social and nonsocial reward moderate the relation between autism symptoms and loneliness in adults with ASD, depression, and controls

    Han GT, Tomarken AJ, Gotham KO. Social and nonsocial reward moderate the relation between autism symptoms and loneliness in adults with ASD, depression, and controls. Autism Res 2019;12(6):884–

  8. [8]

    doi:https://doi.org/10.1002/aur.2088

  9. [9]

    Depression is not a consistent syndrome: an investigation of unique symptom patterns in the STAR*D study

    Fried EI, Nesse RM. Depression is not a consistent syndrome: an investigation of unique symptom patterns in the STAR*D study. J Affect Disord 2015;172:96–102. doi:https://doi.org/10. 1016/j.jad.2014.10.010

  10. [10]

    Detecting depression and mental illness on social media: an integrative review

    Guntuku SC, Yaden DB, Kern ML, Ungar LH, Eichstaedt JC. Detecting depression and mental illness on social media: an integrative review. Curr Opin Behav Sci 2017;18:43–49. doi:https://doi. org/10.1016/j.cobeha.2017.07.005

  11. [11]

    Detecting and measuring depression on social media using a machine learning approach: systematic review

    Liu D, Feng XL, Ahmed F, Shahid M, Guo J. Detecting and measuring depression on social media using a machine learning approach: systematic review. JMIR Ment Health 2022;9(3):e27244. doi: https://doi.org/10.2196/27244

  12. [12]

    Detection of depression-related posts in Reddit social media forum

    Tadesse MM, Lin H, Xu B, Yang L. Detection of depression-related posts in Reddit social media forum. IEEE Access 2019;7:44883–44893. doi:https://doi.org/10.1109/access.2019. 2909180

  13. [13]

    Language use of depressed and depression-vulnerable college students

    Rude SS, Gortner EM, Pennebaker JW. Language use of depressed and depression-vulnerable college students. Cogn Emot 2004;18(8):1121–1133. doi:https://doi.org/10.1080/ 02699930441000030

  14. [14]

    BERT: pre-training of deep bidirectional transformers for language understanding

    Devlin J, Chang MW, Lee K, Toutanova K. BERT: pre-training of deep bidirectional transformers for language understanding. In: Proceedings of NAACL-HLT 2019. 2019:4171–4186. doi:https: //doi.org/10.18653/v1/N19-1423

  15. [15]

    Language of ADHD in adults on so- cial media

    Guntuku SC, Ramsay JR, Merchant RM, Ungar LH. Language of ADHD in adults on so- cial media. J Atten Disord 2019;23(12):1475–1485. doi:https://doi.org/10.1177/ 1087054717738083

  16. [16]

    Using #ActuallyAutistic on Twitter for precision diagnosis of autism spec- trum disorder: machine learning study

    Jaiswal A, Washington P. Using #ActuallyAutistic on Twitter for precision diagnosis of autism spec- trum disorder: machine learning study. JMIR Form Res 2024;8:e52660. doi:https://doi.org/ 10.2196/52660

  17. [17]

    MentalBERT: publicly available pretrained lan- guage models for mental healthcare

    Ji S, Zhang T, Ansari L, Fu J, Tiwari P, Cambria E. MentalBERT: publicly available pretrained lan- guage models for mental healthcare. In: Proceedings of LREC 2022. 2022:7184–7190. doi:https: //doi.org/10.63317/2d7umxifj8wz

  18. [18]

    RoBERTa: a robustly optimized BERT pretraining approach

    Liu Y , Ott M, Goyal N, Du J, Joshi M, Chen D, Levy O, Lewis M, Zettlemoyer L, Stoyanov V . RoBERTa: a robustly optimized BERT pretraining approach. arXiv:1907.11692 [Preprint]. 2019. doi: https://doi.org/10.48550/arXiv.1907.11692

  19. [19]

    ReDSM5: a Reddit dataset for DSM-5 depression detection

    Bao E, Pérez A, Parapar J. ReDSM5: a Reddit dataset for DSM-5 depression detection. In: Proceedings of CIKM ’25. 2025. doi:https://doi.org/10.1145/3746252.3761610

  20. [20]

    Extracting mental health indicators from English and Spanish social media: a machine learning approach

    Villa-Pérez ME, Trejo LA, Moin MB, Stroulia E. Extracting mental health indicators from English and Spanish social media: a machine learning approach. IEEE Access 2023;11:128135–128152. doi: https://doi.org/10.1109/access.2023.3332289 16

  21. [21]

    Diagnostic and Statistical Manual of Mental Disorders

    American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders. 5th ed. Washington, DC: American Psychiatric Publishing; 2013. doi:https://doi.org/10.1176/ appi.books.9780890425787

  22. [22]

    Cross-platform detection of psychiatric hospitalization via social media data: comparison study

    Nguyen VC, Lu N, Kane JM, Birnbaum ML, De Choudhury M. Cross-platform detection of psychiatric hospitalization via social media data: comparison study. JMIR Ment Health 2022;9(12):e39747. doi: https://doi.org/10.2196/39747

  23. [23]

    Building efficient universal classifiers with natu- ral language inference

    Laurer M, van Atteveldt W, Casas A, Welbers K. Building efficient universal classifiers with natu- ral language inference. arXiv:2312.17543. 2023 Dec 29. doi:https://doi.org/10.48550/ arXiv.2312.17543

  24. [24]

    Quantifying changes in the language used around mental health on Twitter over 10 years: observational study

    Stupinski AM, Alshaabi T, Arnold MV , Adams JL, Minot JR, Price M, Dodds PS, Danforth CM. Quantifying changes in the language used around mental health on Twitter over 10 years: observational study. JMIR Ment Health 2022;9(3):e33685. doi:https://doi.org/10.2196/33685

  25. [25]

    The World Federation of ADHD International Consensus Statement: 208 evidence-based conclusions about the disorder

    Faraone SV , Banaschewski T, Coghill D, Zheng Y , Biederman J, Bellgrove MA, et al. The World Federation of ADHD International Consensus Statement: 208 evidence-based conclusions about the disorder. Neurosci Biobehav Rev 2021;128:789–818. doi:https://doi.org/10.1016/j. neubiorev.2021.01.022

  26. [26]

    Shared heritability of attention-deficit/hyperactivity disorder and autism spectrum disorder

    Rommelse NN, Franke B, Geurts HM, Hartman CA, Buitelaar JK. Shared heritability of attention-deficit/hyperactivity disorder and autism spectrum disorder. Eur Child Adolesc Psychiatry 2010;19(3):281–295. doi:https://doi.org/10.1007/s00787-010-0092-x

  27. [27]

    Prevalence and characteristics of autism spectrum disorder among children aged 8 years — Autism and Developmental Disabilities Monitoring Network, 11 sites, United States, 2020

    Maenner MJ, Warren Z, Williams AR, Amoakohene E, Bakian A V , Bilder DA, et al. Prevalence and characteristics of autism spectrum disorder among children aged 8 years — Autism and Developmental Disabilities Monitoring Network, 11 sites, United States, 2020. MMWR Surveill Summ 2023;72(2):1–

  28. [28]

    doi:https://doi.org/10.15585/mmwr.ss7202a1

  29. [29]

    Prevalence of co-occurring mental health diagnoses in the autism population: a systematic review and meta-analysis

    Lai MC, Kassee C, Besney R, Bonato S, Hull L, Mandy W, et al. Prevalence of co-occurring mental health diagnoses in the autism population: a systematic review and meta-analysis. Lancet Psychiatry 2019;6(10):819–829. doi:https://doi.org/10.1016/s2215-0366(19)30289-5

  30. [30]

    Understanding and prevention of suicide in autism

    Cassidy S, Rodgers J. Understanding and prevention of suicide in autism. Lancet Psychiatry 2017;4(6):e11. doi:https://doi.org/10.1016/s2215-0366(17)30162-1

  31. [31]

    Putting on my best normal

    Hull L, Petrides KV , Allison C, Smith P, Baron-Cohen S, Lai MC, Mandy W. “Putting on my best normal”: social camouflaging in adults with autism spectrum conditions. J Autism Dev Disord 2017;47(8):2519–2534. doi:https://doi.org/10.1007/s10803-017-3166-5

  32. [32]

    Sleep problems in adults with ADHD: prevalences and their relationship with psychiatric comorbidity

    van der Ham M, Bijlenga D, Böhmer M, Beekman ATF, Kooij JJS. Sleep problems in adults with ADHD: prevalences and their relationship with psychiatric comorbidity. J Atten Disord 2024;28(9):1228–1240. doi:https://doi.org/10.1177/10870547241284477

  33. [33]

    The risk of eating disorders comorbid with attention-deficit/hyperactivity disorder: a systematic review and meta-analysis

    Nazar BP, Bernardes C, Peachey G, Sergeant J, Mattos P, Treasure J. The risk of eating disorders comorbid with attention-deficit/hyperactivity disorder: a systematic review and meta-analysis. Int J Eat Disord 2016;49(12):1045–1057. doi:https://doi.org/10.1002/eat.22643

  34. [34]

    Major depressive disorder is associated with broad impairments on neuropsychological measures of executive function: a meta-analysis and review

    Snyder HR. Major depressive disorder is associated with broad impairments on neuropsychological measures of executive function: a meta-analysis and review. Psychol Bull 2013;139(1):81–132. doi: https://doi.org/10.1037/a0029301 17

  35. [35]

    From ADHD to SAD: analyzing the language of mental health on Twitter through self-reported diagnoses

    Coppersmith G, Dredze M, Harman C, Hollingshead K. From ADHD to SAD: analyzing the language of mental health on Twitter through self-reported diagnoses. In: Proceedings of the 2nd Workshop on Computational Linguistics and Clinical Psychology. 2015:1–10. doi:https://doi.org/10. 3115/v1/w15-1201

  36. [36]

    The psychological meaning of words: LIWC and computerized text analysis methods

    Tausczik YR, Pennebaker JW. The psychological meaning of words: LIWC and computerized text analysis methods. J Lang Soc Psychol 2010;29(1):24–54. doi:https://doi.org/10.1177/ 0261927x09351676

  37. [37]

    Natural language processing applied to mental illness detection: a narrative review

    Zhang T, Schoene AM, Ji S, Ananiadou S. Natural language processing applied to mental illness detection: a narrative review. NPJ Digit Med 2022;5(1):46. doi:https://doi.org/10.1038/ s41746-022-00589-7

  38. [38]

    Attention is all you need

    Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, Kaiser L, Polosukhin I. Attention is all you need. Adv Neural Inf Process Syst 2017;30:5998–6008. doi:https://doi.org/10. 48550/arXiv.1706.03762

  39. [39]

    Multi-label classification: an overview

    Tsoumakas G, Katakis I. Multi-label classification: an overview. Int J Data Warehous Min 2007;3(3):1–13. doi:https://doi.org/10.4018/jdwm.2007070101

  40. [40]

    Survey on deep learning with class imbalance

    Johnson JM, Khoshgoftaar TM. Survey on deep learning with class imbalance. J Big Data 2019;6(1):27. doi:https://doi.org/10.1186/s40537-019-0192-5

  41. [41]

    Improving neural networks by preventing co-adaptation of feature detectors

    Hinton GE, Srivastava N, Krizhevsky A, Sutskever I, Salakhutdinov RR. Improving neural networks by preventing co-adaptation of feature detectors. arXiv:1207.0580. 2012. doi:https://doi.org/ 10.48550/arXiv.1207.0580

  42. [42]

    Decoupled weight decay regularization

    Loshchilov I, Hutter F. Decoupled weight decay regularization. In: Proceedings of ICLR 2019. 2019. doi:https://doi.org/10.48550/arXiv.1711.05101

  43. [43]

    Regression shrinkage and selection via the lasso

    Tibshirani R. Regression shrinkage and selection via the lasso. J R Stat Soc Series B 1996;58(1):267–

  44. [44]

    doi:https://doi.org/10.1111/j.2517-6161.1996.tb02080.x

  45. [45]

    A study of cross-validation and bootstrap for accuracy estimation and model selection

    Kohavi R. A study of cross-validation and bootstrap for accuracy estimation and model selection. In: Proceedings of IJCAI-95. 1995:1137–1145. doi:https://doi.org/10.5555/1643031. 1643047

  46. [46]

    Bootstrap methods: another look at the jackknife

    Efron B. Bootstrap methods: another look at the jackknife. Ann Stat 1979;7(1):1–26. doi:https: //doi.org/10.1214/aos/1176344552

  47. [47]

    The meaning and use of the area under a receiver operating characteristic (ROC) curve

    Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143(1):29–36. doi:https://doi.org/10.1148/radiology. 143.1.7063747

  48. [48]

    Harnessing smartphone-based digital phenotyping to enhance behavioral and mental health

    Onnela JP, Rauch SL. Harnessing smartphone-based digital phenotyping to enhance behavioral and mental health. Neuropsychopharmacology 2016;41(7):1691–1696. doi:https://doi.org/10. 1038/npp.2016.7

  49. [49]

    Ethical issues in using Twitter for population-level depression moni- toring: a qualitative study

    Mikal J, Hurst S, Conway M. Ethical issues in using Twitter for population-level depression moni- toring: a qualitative study. BMC Med Ethics 2016;17(1):22. doi:https://doi.org/10.1186/ s12910-016-0105-5 18

  50. [50]

    Social data: biases, methodological pitfalls, and ethi- cal boundaries

    Olteanu A, Castillo C, Diaz F, Kıcıman E. Social data: biases, methodological pitfalls, and ethi- cal boundaries. Front Big Data 2019;2:13. doi:https://doi.org/10.3389/fdata.2019. 00013

  51. [51]

    DeBERTaV3: improving DeBERTa using ELECTRA-style pre-training with gradient-disentangled embedding sharing

    He P, Gao J, Chen W. DeBERTaV3: improving DeBERTa using ELECTRA-style pre-training with gradient-disentangled embedding sharing. In: Proceedings of ICLR 2023. 2023. doi:https: //doi.org/10.48550/arXiv.2111.09543

  52. [52]

    Hugging Face

    cross-encoder/nli-deberta-v3-large [Machine learning model]. Hugging Face. URL:https:// huggingface.co/cross-encoder/nli-deberta-v3-large[accessed 2026-07-05]

  53. [53]

    Using text mining for study iden- tification in systematic reviews: a systematic review of current approaches

    O’Mara-Eves A, Thomas J, McNaught J, Miwa M, Ananiadou S. Using text mining for study iden- tification in systematic reviews: a systematic review of current approaches. Syst Rev 2015;4:5. doi: https://doi.org/10.1186/2046-4053-4-5

  54. [54]

    Users with≥1 pre-filtered tweet

    Nogueira R, Cho K. Passage re-ranking with BERT. arXiv:1901.04085 [Preprint]. 2019. doi:https: //doi.org/10.48550/arXiv.1901.04085 Supplementary Material Table S1: Cross-threshold discrimination performance of the L1-penalised logistic classifier.For each swept pre-filter threshold, the number of users retained after the per-tweet symptom-relevance gate a...