Prediction of Graduation Development Based on Hypergraph Contrastive Learning With Imbalanced Sampling

Abstract

With the increasingly competitive job market, the employment issue for college graduates has received more and more attention. Predicting graduation development can help students understand their suitable graduation development, thus easing the pressure of finding employment after graduation. However, existing research must look into the issue of imbalance and long-tail distribution in student graduation development. This paper proposes a novel hypergraph contrastive learning model based on imbalanced sampling (IS-HGCL) that enables us to address this problem. First, construct a hypergraph using students' school performance and social behavior. Then, our proposed imbalanced sampling strategy is applied to optimize the hypergraph structure and alleviate the imbalance issue. A self-updating hypergraph neural network is designed to optimize hyperedge representation and alleviate the long-tail distribution issue to enhance the hypergraph representation further. Finally, the structural consistency between the two optimized hypergraphs is maximized via node-level contrastive learning. Experiments on a real-world campus dataset demonstrate the superiority of the IS-HGCL model.