Traditional cognitive diagnosis models primarily focus on students’ knowledge states while neglecting the impact of emotional factors on learning performance. In recent years, deep learning methods have made progress in dynamic knowledge tracing, yet they have not fully considered the role of emotions. This study proposes the Emotion-Aware Knowledge Tracing (EAKT) model, which integrates multi-head emotional attention and dynamic emotional gating mechanisms to enhance student performance prediction. EAKT independently models the influence of different emotions on knowledge states through the emotional attention mechanism and adjusts knowledge representations via dynamic emotional gating to accommodate emotional fluctuations. Experimental results on the ASSISTments2012 and ASSISTments2017 datasets demonstrate that EAKT outperforms baseline models in ACC, AUC, and RMSE. The findings suggest that incorporating emotional factors improves the accuracy of knowledge tracing and provides a novel theoretical framework for personalized educational interventions.