Knowledge tracing (KT) aims to predict learners’ future performance by modeling their historical interaction data. In real-world educational scenarios, it is difficult to effectively capture hierarchical and semantic relationships between questions and knowledge components (KCs) (A knowledge component (KC) is a generalization of everyday terms like concept, principle, fact, or skill.). Furthermore, real-world datasets are very noisy and how to build a robust deep learning based KT models still faces challenges. To address these problems, we propose a hierarchical contrastive graph framework for KT tasks, i.e., HCGKT, which combines hierarchical graph filtering attention, adversarial contrastive learning, and graph convolutional networks. Experiments on three datasets demonstrate our model’s superior performance in terms of prediction accuracy. To encourage reproducible research, we have made our code publicly available at https://pykt.org/ .