Perceiving distress in digital entertainment spaces: Multimodal linguistic–behavioural deep learning and benchmark datasets for proactive depression detection on social media
Year: 2026
Type: article
Abstract: People increasingly express their emotions and experiences on social media and short-form video platforms, making these spaces valuable sources of multimodal signals for understanding mental health. Depression is among the leading causes of disability worldwide, yet conventional screening methods are often infrequent and rely heavily on self-reported symptoms. This paper investigates the potential of entertainment-oriented social media platforms for proactive depression detection by analysing the emotional and behavioural patterns reflected in users’ online activities. An empirical and reproducible framework is developed using three publicly available Reddit benchmark datasets. Both conventional machine learning and deep learning approaches are evaluated, together with an intermediate fusion framework that integrates contextual textual information with linguistic and behavioural features. After duplicate posts are removed and bootstrap confidence intervals are reported, experimental findings show that TF–IDF-based linear models remain highly effective, while behavioural features provide complementary information but do not consistently improve results when combined with lexical representations, and do not surpass the text baseline at the current dataset scale. Keyword-masking analyses further indicate that detection does not rely on explicit diagnostic terms. The proposed framework provides a reproducible foundation for future extensions toward richer multimodal depression detection systems.
Source: Systems and Soft Computing
Authors: Sabahat Nazir, Hilal Ahmad Khanday
Institution: University of Kashmir
Language: English
Topic: Mental Health via Writing
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $1,870
OpenAlex ID: W7225264631
