AI framework tracks depression shifts from user social media posts
BERT models + temporal analysis spot mental health changes better than LLMs alone
A team of researchers (Belcastro, Gervino, Marozzo, Talia, Trunfio) have introduced a novel explainable AI framework designed to detect shifts in depression status from users' digital traces—such as social media posts, chats, and online interactions. The system organizes these timestamped traces into temporal trajectories that capture how mental health signals evolve over time, identifying phases of improvement, deterioration, or stability. At its core, the framework combines multiple BERT-based models to extract complementary signals across different dimensions: sentiment, emotion, and depression severity. These signals are then aggregated and analyzed to pinpoint meaningful change points. To boost interpretability, a large language model (LLM) generates concise, human-readable reports that describe the evolution of mental health signals and highlight key transitions.
In evaluation on two social media datasets, the approach produced more coherent and informative summaries than direct LLM-based reporting, achieving higher coverage of user history, stronger temporal coherence, and improved sensitivity to change points. An ablation study confirmed that each component—especially temporal modeling and segmentation—contributes significantly to performance. The method provides an interpretable, non-clinical view of mental health signals over time, supporting research and decision-making without aiming at clinical diagnosis. This work represents a step forward in using AI to understand mental health from the digital footprints users leave behind.
- Framework uses multiple BERT-based models to extract sentiment, emotion, and depression severity signals from digital traces.
- Temporal trajectories and change point analysis detect shifts in mental state; LLM generates explainable reports.
- Outperforms direct LLM-based reporting on two social media datasets, with ablation confirming critical role of temporal modeling.
Why It Matters
Enables researchers to non-invasively track mental health trends from social data using interpretable AI, without clinical diagnosis.