A Multi-Task Learning Framework for Balancing Mental Health and Social Media Usage using Digital Behavioural Data

The fast advancement in technology is constantly changing how people live and having a greater impaction than ever before. Various human emotional changes are largely associated with the evolution of new technology. Instead of treating these changes differently, the contemporary methods predict methods one at a time, this can lead to the loss of common triggers across them. This kind of partitioned approach can lead to the overall weaker predictions. This approach moves ahead by combining those tasks into a single system that learns them all together by using daily life features. Using information like how someone sleeps, their screen habits, and productivity throughout the day, it is found that neural networks with attention mechanisms can predict multiple tasks simultaneously. Each task focuses on its own attention mechanisms that allow it to pick out essential aspects of the latent representation and ignoring noise from the non-relevant tasks’ latent representations. Such system can predict numerical outcomes connected to mental wellness, attention, and screen dependence and even predict binary classifications connected with the detection of mental health concerns. Each of these networks works as intended, with strong results in predicting attention and screen usage patterns and binary classifications regarding mental health concerns. By training these networks on multiple tasks simultaneously through multi-task learning approaches with attention mechanisms focusing on individual tasks’ needs, stronger results emerge as compared to state-of-the-art single-task learning models. Compared to baseline and ablation studies, this multi-task attention learning model outperforms less complicated approaches. Hence, there is significant proof from the test set to show the strength of multi-task learning with attention focused on individual tasks when analyzing aspects like general view mental wellness. What stands out is that the system built here could grow easily, fitting into digital tools meant to track mental health while spotting risks earlier and tailoring insights to individuals.