Student mental health issues are becoming increasingly severe, yet traditional scale assessments suffer from limitations such as high subjectivity and delayed feedback.To address this challenge, this paper proposes an intelligent evaluation framework multi-branch adaptive social-emotional fusion network that integrates social-emotional analysis with multi-branch neural networks.This framework continuously and seamlessly integrates multidimensional digital footprints generated by students in campus life (e.g., text, voice, behavioural patterns) to enable dynamic psychological risk assessment.In this work, digital footprints refer to passively collected multimodal data from students' daily digital activities, including text messages, voice recordings, smartphone sensor logs, and social interaction records.Experiments on the public studentlife dataset demonstrate that the proposed method achieves a key evaluation metric area under the receiver operating characteristic curve of 0.927, surpassing mainstream unimodal models by over 3.2%.It also achieves an overall accuracy of 89.7% and passes statistical significance tests.This confirms the effectiveness and feasibility of utilising multi-source socio-emotional signals for early, objective intervention.
Yue Zheng (Thu,) studied this question.