MTDRA-Net: graph-temporal neural network for dynamic evaluation of college students' responsibility

Responsibility awareness is a core psychological trait in higher education, yet traditional scale assessments suffer from low stability and limited structural interpretability. This study proposes the Multi-Task Dynamic Responsibility Awareness Network (MTDRA-Net), integrating graph attention and temporal convolution networks. Using longitudinal data from 244 undergraduates, the model captures asymmetric inter-dimensional interactions and heterogeneous temporal dynamics across cognition, emotion, volition, and behavior. Results show MTDRA-Net achieves an intra-class correlation coefficient of 0.90, outperforming traditional scales by 12.5%. Confirmatory factor analysis confirms superior structural validity (CFI = 0.955, RMSEA = 0.05). This model was designed to provide a more stable and structurally consistent assessment tool for evaluating college students' sense of responsibility while using the same measurement instrument. It also provided technical support for teachers and administrators in making continuous student development decisions.