SleepDepNet: A multi-task transformer model for assessing sleep quality and depression risk from social media narratives.

Résumé

The relationship between sleep and mental health, particularly depression, represents a critical area of study with significant implications for individual well-being and public health. This work introduces SleepDepNet, a transformer-based multi-task learning framework designed to jointly model sleep quality and depressive sentiment from user-generated text. Using data collected from Reddit communities related to sleep and mental health, the proposed approach integrates attention mechanisms, emotion-aware features, and topic modelling to capture nuanced linguistic patterns associated with sleep disturbances and emotional states. Experimental results demonstrate that SleepDepNet outperforms baseline models, achieving F1-scores of 0.89 for sleep quality classification and 0.86 for depressive sentiment analysis. The model's attention mechanisms provide interpretability by highlighting linguistically salient indicators linked to emotional and sleep-related expressions. Additionally, the proposed SleepDepScore, which integrates outputs from both tasks, offers a unified measure for assessing combined risk levels and supporting prioritization in downstream applications. Overall, the findings suggest that multi-task learning offers a promising direction for modelling complex relationships between sleep and mental health in online discourse. While the results demonstrate strong performance under controlled conditions, the framework is designed to support scalable analysis and can inform future research on digital mental health monitoring and early risk identification.