Framelet Based Dual Hypergraph Neural Networks for Student Engagement Prediction

Ming Li, Shijiandong

Track: Innovations in AI for Education (Day 1)

Paper Length: short-paper (2 pages + references)

Keywords: Student Engagement Prediction, Hypergraph Neural Networks, Framelet Transform

TL;DR: A novel approach using framelet transforms and dual hypergraph neural networks to improve the prediction of student engagement through the analysis of visual features and high-order relationships.

Abstract: In this short (work-in-progress) paper, we focus on the critical task of predicting student engagement. We introduce an innovative framelet transform, designed to proficiently convert students' visual features into sets of low-pass and high-pass coefficients. By placing specific emphasis on these coefficients, we create diverse hypergraphs that capture high-order relationships among students at varying scales. Subsequently, we develop dual hypergraph neural networks to effectively learn these hypergraphs, discerning the unique contributions of low-pass and high-pass components. Preliminary experimental findings on a real-world educational dataset highlight the promising potential of our framework in advancing student engagement prediction models.

Submission Number: 22