Knowledge tracing is the task of learning the growth process of students’ knowledge level by predicting how well the student will answer the next question. Most existing models often fail to account for the relevance between the student’s prior question history and the upcoming question. The datasets they employed are insufficient to predict students’ problem-solving outcomes. However, conventional data augmentation methods fail to address this limitation. To resolve this limitation, we propose a novel model that contains a comparison module and a similarity-driven data augmentation module. These two modules can identify students whose question histories are irrelevant to the target question and select similar students for data augmentation. Extensive experiments conducted on three benchmark datasets demonstrate that our proposed comparison module and data augmentation module achieve statistically significant improvements over existing approaches. Further analysis highlights the specific contributions of our modules to improving knowledge tracing performance.