Harnessing multimodal graph neural networks to predict graduate employment anxiety

This study tackles the crucial problem of forecasting employment anxiety in graduates through surpassing the constraints of traditional evaluation techniques that depend on static self-reports and ignore the dynamic social context.We present a multimodal dynamic graph neural network framework which fuses Smartphone-sensed behavioural data and real-world physical interaction networks.We use a bidirectional gated recurrent unit to encode the temporal behaviour, and we also have a dynamic graph attention to capture the changes in social support.Comprehensive evaluation on the StudentLife dataset shows that our approach outperforms state-of-the-art baselines significantly by reducing root mean square error by 12.5% from 5.82 to 5.09 and mean absolute error by 15.8% from 4.68 to 3.94 and getting a determination coefficient of 0.67.These results show that the framework works well at copying how individual behaviours and social forces work together for mental health checks.