{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,6]],"date-time":"2026-10-06T00:19:11Z","timestamp":1791245951150,"version":"4.1.0"},"reference-count":21,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,3,23]],"date-time":"2026-03-23T00:00:00Z","timestamp":1774224000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Digit. Health"],"abstract":"<jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>Social media provides timely proxy signals of mental health, but reliable tweet-level classification of depression subtypes remains challenging due to short, noisy text, overlapping symptomatology, and labeling bias. Large language models (LLMs) are increasingly used in mental health for tasks such as symptom extraction, risk screening, and triage, yet their reliability for fine-grained depression subtype classification from brief social media posts remains underexplored.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Objective<\/jats:title>\n                    <jats:p>We benchmarked few-shot, prompt-only LLMs against parameter-efficient fine-tuned encoders for identifying depression subtypes in posts on X (formerly Twitter).<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>We used a curated dataset of 14,983 English-language tweets stratified into six clinically grounded categories: five depression subtypes (postpartum, major, bipolar, psychotic, atypical) and a no-depression class. We compared (i) instruction-tuned causal LLMs in a few-shot setting and (ii) supervised fine-tuning of transformer encoders (e.g., RoBERTa, DeBERTa, BERTweet) under identical splits and metrics. The primary evaluation metric was macro-F1 (with accuracy, precision, recall as secondary). We also report per-class precision, recall, and F1 scores, along with confusion matrices, for the best-performing model from each model family.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>Few-shot LLMs achieved macro-F1\u2009=\u20090.73\u20130.77 (best: Llama-3-8B, accuracy 0.75). Fine-tuned encoders consistently outperformed prompt-only models, reaching macro-F1\u2009=\u20090.94\u20130.96 (best: RoBERTa-large, accuracy 0.954). Relative improvements were largest for the clinically challenging classes. Fine-tuning increased F1 for postpartum and psychotic subtypes to \u22480.99 (substantially above few-shot) and boosted major-depression recall from \u22480.53\u20130.60 to \u22480.95\u20130.97. Error analyses showed prompt-only models frequently misclassified major and atypical depression as bipolar, patterns substantially reduced by fine-tuning.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusions<\/jats:title>\n                    <jats:p>On tweet-level depression subtyping, task-specific adaptation via fine-tuning yields substantially higher and more stable performance than few-shot prompting, particularly for nuanced, clinically anchored classes. These findings recommend fine-tuned encoders as strong, compute-efficient baselines for depression subtype classification from social media.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/fdgth.2026.1790533","type":"journal-article","created":{"date-parts":[[2026,3,23]],"date-time":"2026-03-23T10:31:01Z","timestamp":1774261861000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":4,"title":["Depression subtype classification from social media posts: few-shot prompting vs. fine-tuning of large language models"],"prefix":"10.3389","volume":"8","author":[{"given":"Rawan","family":"AlSaad","sequence":"first","affiliation":[{"name":"Weill Cornell Medicine-Qatar","place":["Doha, Qatar"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sulaiman","family":"Alshakhs","sequence":"additional","affiliation":[{"name":"Weill Cornell Medicine-Qatar","place":["Doha, Qatar"]},{"name":"Mental Health Services, Hamad Medical Corporation","place":["Doha, Qatar"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Rajat","family":"Thomas","sequence":"additional","affiliation":[{"name":"Weill Cornell Medicine-Qatar","place":["Doha, Qatar"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,3,23]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"360","DOI":"10.1016\/j.jad.2020.08.091","article-title":"A systematic review of the validity of screening depression through Facebook, Twitter, Instagram, and Snapchat","volume":"286","author":"Kim","year":"2021","journal-title":"J Affect Disord"},{"key":"B2","doi-asserted-by":"publisher","first-page":"4709","DOI":"10.1007\/s11227-021-04040-8","article-title":"Automatic detection of depression symptoms in twitter using multimodal analysis","volume":"78","author":"Safa","year":"2022","journal-title":"J Supercomput"},{"key":"B3","doi-asserted-by":"publisher","first-page":"102716","DOI":"10.1016\/j.artmed.2023.102716","article-title":"Depression detection for twitter users using sentiment analysis in English and Arabic tweets","volume":"147","author":"Helmy","year":"2024","journal-title":"Artif Intell Med"},{"key":"B4","doi-asserted-by":"publisher","first-page":"17","DOI":"10.1517\/14656566.2016.1104299","article-title":"The use of antidepressants in bipolar disorder patients with depression","volume":"17","author":"Bowden","year":"2016","journal-title":"Expert Opin Pharmacother"},{"key":"B5","doi-asserted-by":"publisher","first-page":"1663","DOI":"10.1016\/S0140-6736(13)60989-7","article-title":"Bipolar disorder diagnosis: challenges and future directions","volume":"381","author":"Phillips","year":"2013","journal-title":"Lancet"},{"key":"B6","first-page":"40","volume-title":"Challenges in Differentiating and Diagnosing","author":"Rothschild","year":"2006"},{"key":"B7","author":"Nguyen","year":"2020"},{"key":"B8","doi-asserted-by":"crossref","DOI":"10.1109\/ICCIT60459.2023.10441344","article-title":"DARN: dual-attention RoBERTa network for depression severity detection from twitter","author":"Zaman","year":"2023"},{"key":"B9","doi-asserted-by":"publisher","first-page":"107503","DOI":"10.1016\/j.chb.2022.107503","article-title":"DEPTWEET: a typology for social media texts to detect depression severities","volume":"139","author":"Kabir","year":"2023","journal-title":"Comput Human Behav"},{"key":"B10","doi-asserted-by":"crossref","DOI":"10.18653\/v1\/2020.coling-main.61","article-title":"Identifying depressive symptoms from Tweets: figurative language enabled multitask learning framework","author":"Yadav","year":"2020"},{"key":"B11","first-page":"302","article-title":"AraDepSu: detecting depression and suicidal ideation in Arabic Tweets using transformers","author":"Hassib","year":"2022"},{"key":"B12","first-page":"672","article-title":"Diverse perspectives, divergent models: cross-cultural evaluation of depression detection on Twitter","author":"Abdelkadir","year":"2024"},{"key":"B13","doi-asserted-by":"crossref","first-page":"155","DOI":"10.54941\/ahfe1001458","article-title":"Detecting potential depressed users in twitter using a fine-tuned DistilBERT model","volume-title":"Artificial Intelligence and Social Computing","author":"Adarlo","year":"2022"},{"key":"B14","author":"Nusrat","year":"2024"},{"key":"B15","author":"Nusrat","year":"2024"},{"key":"B16","volume-title":"Diagnostic and Statistical Manual of Mental Disorders","year":"2013"},{"key":"B17","doi-asserted-by":"publisher","first-page":"468","DOI":"10.1007\/s11920-014-0468-6","article-title":"Perinatal depression: an update and overview","volume":"16","author":"Stuart-Parrigon","year":"2014","journal-title":"Curr Psychiatry Rep"},{"key":"B18","doi-asserted-by":"publisher","first-page":"76","DOI":"10.1016\/j.jad.2015.06.020","article-title":"Atypical features in depression: association with obesity and bipolar disorder","volume":"185","author":"\u0141ojko","year":"2015","journal-title":"J Affect Disord"},{"key":"B19","doi-asserted-by":"publisher","first-page":"101","DOI":"10.1146\/annurev-publhealth-040119-094402","article-title":"Social media- and internet-based disease surveillance for public health","volume":"41","author":"Aiello","year":"2020","journal-title":"Annu Rev Public Health"},{"key":"B20","doi-asserted-by":"publisher","first-page":"e69284","DOI":"10.2196\/69284","article-title":"The applications of large language models in mental health: scoping review","volume":"27","author":"Jin","year":"2025","journal-title":"J Med Internet Res"},{"key":"B21","doi-asserted-by":"publisher","first-page":"67","DOI":"10.35566\/jbds\/caoyc","article-title":"Machine learning approaches for depression detection on social Media: a systematic review of biases and methodological challenges","volume":"5","author":"Cao","year":"2025","journal-title":"J Behav Data Sci"}],"container-title":["Frontiers in Digital Health"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fdgth.2026.1790533\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,23]],"date-time":"2026-03-23T10:31:03Z","timestamp":1774261863000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fdgth.2026.1790533\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,3,23]]},"references-count":21,"alternative-id":["10.3389\/fdgth.2026.1790533"],"URL":"https:\/\/doi.org\/10.3389\/fdgth.2026.1790533","relation":{},"ISSN":["2673-253X"],"issn-type":[{"value":"2673-253X","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,3,23]]},"article-number":"1790533"}}