Is it possible to detect suicide risk on social media? A systematic review of psychologically informed machine learning models
Frontiers in Psychology · 04.11.2026 · 17
DOI: 10.3389/fpsyg.2026.1731732 ↗Kurz gefasst
Systematische Übersicht in der Zeitschrift Frontiers in Psychology, online erschienen am 04.11.2026. Verfasst von Claudia García-Martínez und 3 weiteren Autor:innen. Beteiligte Einrichtung: Department of Psychology and Sociology, University of Zaragoza, Zaragoza, Spain. Die Arbeit nennt 56 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Introduction
Suicide is a global health priority, particularly among young people. This systematic review aimed to synthesize current evidence on whether social media data can be used to identify suicide-related risk through psychologically informed machine learning models.
Methods
PubMed, Web of Science, Scopus, ScienceDirect, and IEEE Xplore were searched for articles published between 2024 and 2026, updating previous searches conducted in July 2021, December 2022, and July 2024. Eligible studies used social media data to detect suicidal thoughts and behaviors (STBs) or suicide risk and incorporated psychologically significant variables or representations into the prediction process. Risk of bias and applicability were assessed using the Prediction model Risk of Bias ASsessment Tool for Artificial Intelligence (PROBAST+AI).
Results
A total of 18 studies were included. Models incorporated psycholinguistic and affective features, multiple psychological constructs, and, more recently, temporal, relational, and multimodal information. Commonly represented constructs included depression, anxiety, hopelessness, loneliness, perceived burdensomeness, and emotional states. Reported performance varied considerably across studies, with some models achieving area under the curve (AUC) values above 0.90; however, 13 of the 18 studies were judged to have a high overall risk of bias and five an unclear risk, with concerns concentrated particularly in the analysis domain.
Discussion
Psychologically informed machine learning models show potential for identifying STBs and suicide-related risk from social media data. However, substantial methodological heterogeneity, limited external validation, inconsistent definitions of suicidality, and applicability concerns limit the interpretation and generalizability of current findings. Future research should prioritize externally validated and calibrated models that clearly distinguish between different suicide-related outcomes when assessing individual suicide risk. Systematic review registration https://www.crd.york.ac.uk/PROSPERO/ , identifier CRD42020218349
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Psychology and Sociology, University of Zaragoza, Zaragoza, Spain.
- Institute of Health Research of Aragón (IIS Aragón), Zaragoza, Spain.
- Research Network on Chronicity, Primary Care and Health Promotion RD24/0005/0004, RICAPPS, Madrid, Spain.
- Department of Psychology and Sociology, Faculty of Human Sciences and Education, University of Zaragoza, Huesca, Spain.
Zitieren
APA
García-Martínez, C., López-Del Hoyo, Y., Aguilar-Latorre, A., & Oliván-Blázquez, B. (2026). Is it possible to detect suicide risk on social media? A systematic review of psychologically informed machine learning models. Frontiers in Psychology, 17. https://doi.org/10.3389/fpsyg.2026.1731732
Vancouver
García-Martínez C, López-Del Hoyo Y, Aguilar-Latorre A, Oliván-Blázquez B. Is it possible to detect suicide risk on social media? A systematic review of psychologically informed machine learning models. Frontiers in Psychology. 2026;17. doi:10.3389/fpsyg.2026.1731732
BibTeX
@article{GarcaMartnez2026,
author = {García-Martínez, Claudia and López-Del Hoyo, Yolanda and Aguilar-Latorre, Alejandra and Oliván-Blázquez, Bárbara},
title = {Is it possible to detect suicide risk on social media? A systematic review of psychologically informed machine learning models},
journal = {Frontiers in Psychology},
year = {2026},
volume = {17},
doi = {10.3389/fpsyg.2026.1731732}
}
EUVIMED