Prerequisite Learning through Diversity-driven Selective Attention

ACL ARR 2025 May Submission7355 Authors

Abstract: Prerequisite relation extraction aims to identify concept dependencies, which are crucial for curriculum planning and adaptive education. Existing methods struggle with noisy edges, dense graphs, or fail to model diverse concept relations effectively. In this paper, we propose DPPNet, a novel graph-based approach that incorporates a Determinantal Point Process (DPP) to perform diversity-driven neighbor selection, enabling the model to retain informative and structurally diverse relations while discarding redundancy. Our method integrates this pruning mechanism into the learning pipeline and operates in a single pass, leading to a highly efficient and robust model. Empirical results across three benchmark datasets demonstrate that DPPNet outperforms existing state-of-the-art methods across three key dimensions: classification performance (Accuracy and F1-score), memory footprint, and training time. These results highlight DPPNet’s effectiveness and scalability, making it a practical choice for real-world educational applications.

Paper Type: Long

Research Area: Information Extraction

Research Area Keywords: Prerequisite Relation Extraction, Relation Extraction, Graph Neural Networks

Contribution Types: NLP engineering experiment, Approaches low compute settings-efficiency

Languages Studied: English

Submission Number: 7355