Understanding School Attendance Through Multimodal Modelling of Student Narratives

Tingrui Qiao, Caroline Walker, Chris W Cunningham, Adam Jang-Jones, Susan Mary Bennett Morton, Kane Meissel, Yun Sing Koh

Track: Full paper

Keywords: School Attendance, Computational Social Science, Multimodal Deep Learning

Abstract: Regular school attendance is critical for young people, supporting academic achievement, social development, and the cultivation of lifelong habits. Traditional methods for analyzing attendance patterns often rely on structured survey data targeted to their parents and teachers, which can overlook students' perspectives and experiences. To address this gap, Our Voices programme developed and deployed the Our Journey app to participants of Growing Up in New Zealand, the country’s largest longitudinal study of youth well-being. The app enables young people to share their experiences through multimodal responses, offering unique insights into the factors influencing school attendance. The app data is linked to official attendance records from the Ministry of Education, allowing us to model school attendance outcomes based on students' perspectives. To analyse the data, we propose Thematic Bottleneck Models (TBMs) and integrate them into our attendance modelling framework to enhance the understanding of subjective experiences behind data and the interpretability of predictions. TBMs introduce qualitative concepts as intermediate labels by mapping multimodal data to qualitative insights from thematic analysis before the prediction tasks. Our attendance modelling framework outperforms existing multimodal methods in predicting attendance percentage and persistent absenteeism. Analysis of themes within TBMs reveals distinct motivational and contextual factors associated with regular attendance and persistent absenteeism, highlighting the complex interplay of personal, social, and systemic influences on school engagement. The findings from this study offer valuable insights into the factors influencing school attendance and are being used to inform education policy and guide strategies to support student engagement in New Zealand.

Submission Number: 26