Teacher's mental health risk assessment based on multi-modal data

In recent years, mental health among primary and secondary school teachers has gained significant attention. Teachers' mental well-being is crucial for their professional growth, personal development, and impacts students' progress and societal development. Depression, a prevalent psychological disorder, poses a serious threat to teachers' careers and teaching quality. Therefore, developing effective depression risk assessment methods is urgent. This study focuses on constructing a depression risk assessment model for primary and secondary school teachers using multimodal data to provide a scientific basis for early interventions. The research is divided into three parts: constructing a depression risk dataset, extracting and fusing features from different modal data, and building and validating the model. The study recruited 75 teachers through an experimental platform and the snowball sampling method, collecting text, audio, and facial expression data over 15 consecutive working days using video diaries designed based on Beck's cognitive theory of depression. The model labels were based on scores from the CES-D and SWLS scales. After cleaning and excluding irregular recordings, 1,140 diary data were obtained and divided into training and validation sets. Model construction involved using RoBERTa for text, Mel-ResNet for audio, and 3D-ResNet for facial expressions. The Transformer architecture was used for multimodal fusion, and a classification network was utilized for the final depression risk assessment. Expected results show that multimodal fusion models outperform unimodal models in assessment accuracy and robustness. This study theoretically expands the research perspective on depression risk assessment and validates the effectiveness of multimodal technology. Practically, it provides an innovative tool for precise identification of depression risk among teachers, facilitating early psychological interventions. The method holds promise for broader application to diverse populations and scenarios, bringing new opportunities to mental health assessment.