Depressive tendency, a precursor to major depressive disorder (MDD), is frequently undetected by existing early-warning systems. Traditional methods, based on univariate metrics or subjective inventories, neglect the multidimensional interactions of biomarkers essential for accurate diagnosis. This study aims to enhance the accuracy of depressive tendency detection, thereby contributing to the reduction of MDD incidence and alleviation of its societal burden through early warning and exploratory intervention use.
This study proposed a multidimensional model for detecting depressive tendency, integrating emotional, behavioral, cognitive, physiological and temporal features into a contextual modeling framework. The model combines bidirectional long short-term memory with an attention mechanism to improve detection accuracy and robustness. It has also been integrated into a virtual psychological intervention platform to support user interaction and exploratory intervention use.
The proposed model effectively captures complex manifestations of depressive tendency, including emotional fluctuations, behavioral dynamics and physiological patterns, particularly in social media data. It outperforms existing models in detection accuracy. An exploratory study within a virtual psychological intervention platform demonstrates the model’s practical utility for early detection and intervention.
This study integrates multidimensional psychological and physiological signals into a contextual modeling framework, overcoming limitations of traditional MDD detection methods. The proposed model enhances early detection accuracy and generalizes well across social media datasets. Its practical utility in virtual intervention use highlights its potential for proactive and personalized mental health support.
