[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-81563-en":3,"doc-seo-81563-105":29,"detail-sidebar-cat-0-en-105":90},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},81563,549758146520,"Patrick","https://ap-avatar.wpscdn.com/avatar/80002397d8c0411e94?_k=1775819394049821470",8,"Research & Report","Probabilistic Textual Time Series Depression Detection","Accurate, interpretable depression severity prediction is crucial for clinical decision support, yet many text models offer no uncertainty estimates or temporal interpretability. PTTSD (Probabilistic Textual Time Series Depression Detection) predicts PHQ-8 from clinical interview utterance sequences while producing calibrated uncertainty. The framework supports both seq-to-one and seq-to-seq variants, built from LSTMs, self-attention, and residual connections, with Gaussian or Student’s-t output heads trained via negative log-likelihood. Sequence-to-sequence enables analysis of how confidence evolves during an interview. Evaluations on E-DAIC and DAIC-WOZ show competitive MAE and well-calibrated prediction intervals.","Probabilistic Textual Time Series Depression Detection  \nFabian Schmidt1 , Seyedehmoniba Ravan2 , Vladimir Vlassov1  \n1Department of Computer Science, KTH Royal Institute of Technology, Sweden  \n2Department of Information Technology, Uppsala University, Sweden  \nCorrespondence: [fschm@kth.se](fschm@kth.se)  \narXiv :2511 .04476v2 [ cs .CL] 10 Jul 2026  \nAbstract  \nAccurate and interpretable predictions of depression severity are essential for clinical decision support, yet existing models often lack uncertainty estimates and temporal interpretability. We propose PTTSD, a Probabilistic framework for Depression Detection from clinical interview utterance sequences that predicts PHQ-8 scores while modeling calibrated uncertainty. PTTSD includes sequence-to-sequence and sequence-to-one variants, both combining LSTMs, self-attention, and residual connections with Gaussian or Student’s-t output heads trained via negative log-likelihood. The sequence-to-sequence variant enables temporal analysis of how predictive confidence evolves over an interview, despite the target being a single session-level score. Evaluated on E-DAICand DAIC-WOZ, PTTSD achieves competitive performance among text-only systems (e.g., MAE = 3.85 on E-DAIC, 3.55 on DAIC) and produces well-calibrated prediction intervals.  \nAblations confirm the value of attention and probabilistic modeling, while a three-part calibration analysis and qualitative case studies highlight the clinical relevance of uncertaintyaware prediction.  \n1 Introduction  \nDepression remains one of the leading causes of global disability, affecting over 300 million individuals worldwide (WHO, 2017, 2022) . Scalable, automated tools for assessing depressive symptom severity offer valuable support in digital therapy and remote care, where access to clinicians is limited. Among these tools, text-based systems that process clinical interviews have shown strong potential for predicting standardized scores such asthe PHQ-8 (Kroenke et al., 2009) .  \nRecent methods typically model interview transcripts as sequences of utterances and employ architectures such as LSTMs, Transformers, or large language models (LLMs) (Mandal et al., 2025 ; Fang  \net al., 2023a ; Nykoniuk et al., 2025 ; Sadeghi et al., 2024) . However, most existing approaches produce scalar severity estimates without quantifying uncertainty, which is an important limitation in high-stakes clinical contexts where a prediction of“PHQ-8 = 12” is far more actionable when accompanied by a measure of confidence.  \nWe argue that the sequential nature of clinical interviews creates a natural opportunity to address this gap. Each utterance provides a contextdependent observation, and the cumulative sequence progressively constrains the space ofplausible severity estimates. While the prediction target, i.e., the PHQ-8 score, is a single-session-level value rather than a time-varying quantity, the input is inherently sequential, and modeling it as such offers two key advantages that point-estimate systems forgo. First, it allows probabilistic models to capture aleatoric uncertainty, that is, the uncertainty arising from sparse, contradictory, or ambiguous language, and to express how that uncertainty resolves as context accumulates. Second, it enables interpretable temporal analyses: identifying which utterances drive prediction shifts, and how model confidence stabilizes (or fails to stabilize) over the course of an interview.  \nWe introduce PTTSD, a Probabilistic Textual Time Series Depression Detection model that makes temporally grounded, calibrated predictions over PHQ-8 scores from utterance-level sequences. PTTSD addresses two key gaps in the field. First, it replaces point predictions with calibrated distributional outputs (Gaussian or Student’s-t heads trained via negative log-likelihood), enabling clinicians to assess both predicted severity and the model’s confidence in that prediction. Second, through its sequence-to-sequence (seq-to-seq) v","cbCairvZVihk99t1","https://ap.wps.com/l/cbCairvZVihk99t1","pdf",655542,1,16,"English","en",105,"# Introduction\n# Related Work","[{\"question\":\"What does PTTSD predict, and from what input format?\",\"answer\":\"PTTSD predicts PHQ-8 depression severity from sequences of clinical interview utterances. It treats the interview transcript as a textual time series and learns from utterance-level order.\"},{\"question\":\"How does PTTSD provide uncertainty along with severity scores?\",\"answer\":\"PTTSD uses probabilistic output heads (Gaussian or Student’s-t) trained with negative log-likelihood, producing calibrated prediction intervals rather than only point estimates.\"},{\"question\":\"What is the benefit of the seq-to-seq variant in PTTSD?\",\"answer\":\"The seq-to-seq formulation supports temporal analysis of how predictive confidence changes across the interview. This helps identify which utterances drive prediction shifts and where ambiguity persists.\"}]",1784174348,40,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":27},"probabilistic-textual-time-series-depression-detection","",{"@graph":35,"@context":84},[36,53,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/probabilistic-textual-time-series-depression-detection/81563/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"What does PTTSD predict, and from what input format?","Question",{"text":74,"@type":75},"PTTSD predicts PHQ-8 depression severity from sequences of clinical interview utterances. It treats the interview transcript as a textual time series and learns from utterance-level order.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does PTTSD provide uncertainty along with severity scores?",{"text":79,"@type":75},"PTTSD uses probabilistic output heads (Gaussian or Student’s-t) trained with negative log-likelihood, producing calibrated prediction intervals rather than only point estimates.",{"name":81,"@type":72,"acceptedAnswer":82},"What is the benefit of the seq-to-seq variant in PTTSD?",{"text":83,"@type":75},"The seq-to-seq formulation supports temporal analysis of how predictive confidence changes across the interview. This helps identify which utterances drive prediction shifts and where ambiguity persists.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,118,121,126,129,133],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":28,"slug":117},7,"Healthcare","healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":119,"slug":120},30,"research-report",{"id":122,"doc_module":4,"doc_module_name":45,"category_name":123,"show_sort_weight":124,"slug":125},9,"Religion & Spirituality",20,"religion-spirituality",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":124,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]