Analyzing Social Media to Infer Mental Health Status and Affective States for Crisis and Disaster Management: Scoping Review.

Background: The use of social media (SoMe) during crisis and disaster situations (CaDs) has gained increasing attention across disciplines. However, existing research is highly fragmented and often focused on technical aspects, with a limited understanding of how and which psychosocial information is derived from SoMe in CaDs. Objective: This scoping review provides an overview of the current research landscape regarding the analysis of SoMe data during CaDs to obtain information about public mental health and psychosocial needs. It identifies key themes, methodological approaches, and research gaps, with a particular focus on relevance for the German context. Methods: Following a scoping review protocol, a structured database search was conducted in PubMed, Web of Science, and Scopus to identify peer-reviewed studies published up to 2025. A method of triangulation combining qualitative and quantitative approaches was applied. The studies were analyzed regarding the type of CaDs, geographical focus, classification systems, methods of analysis used, and inclusion of psychosocial aspects (such as affect and mental health status). Results: Overall, we identified 179 studies that examined 267 CaDs. Of the included studies, 76% (136/179) focused on natural disasters, with biological CaDs representing 23% (41/179) of these events. For Germany, 5 studies were found, with only one covering storms, floods, or extreme temperatures, despite these making up most of the disasters in Germany per EM-DAT (Emergency Events Database) data. Most studies used datasets from Asia (especially China), while Africa was examined less often, pointing to differences in geographical representativeness. To infer mental health status or affective state, 47 studies used machine learning, 87 studies used lexicon-based approaches, and 25 studies used a combination; 14 studies used manual coding, and few studies did not explicitly mention their approach. Mental health outcomes ranged from affective valence (positive, negative, or neutral) to specific primary (eg, fear) and secondary (eg, denial) emotions and needs (eg, resources). Yet, few studies were based on theoretical models or included end-user perspectives. No study conducted real-time analysis; instead, all were retrospective. Additionally, current research focuses primarily on deficits (eg, psychological needs, negative affect, or stress), and often neglects positive mental health outcomes (eg, resilience and collective coping). Conclusions: This scoping review underlines the rising popularity of SoMe analysis in CaDs regarding public mental health and needs. Although different techniques were developed and tested, there remain major gaps in real-time application, end-user integration, and contextual adaptation-particularly for underrepresented regions such as Africa, but also in countries such as Germany. As most models were developed or tested retrospectively (eg, using data from the COVID-19 pandemic), future research should examine the validity and tenability of such models in real-time monitoring and data, and emphasize more user-centered design and participatory research, theoretical grounding, and practical utility.