Knowledge Tracing (KT) is a fundamental task in personalized learning to dynamically model the evolution of a student’s knowledge state. However, a key limitation of most existing KT approaches is that they treat knowledge concepts as independent entities, overlooking the interconnected relationships among them. Despite recent progress in graph-based methods that model the knowledge structure, most still neglect to distinguish between different relationship types and to model how knowledge states propagate dynamically. Therefore, in this paper, we propose CGKT, which is a spatio-temporal cognitive graph-enhanced framework for KT. It employs a Gated Recurrent Unit (GRU) to capture temporal learning patterns by updating the knowledge state of the specific skill a student has just practiced. Furthermore, this update triggers a cognitive propagation module that models the spatial influence among knowledge concepts, which decomposes knowledge propagation into three distinct types: Prerequisite, Associative, and Consolidation. The above multi-type spatial influence embeddings will be dynamically integrated through an attention mechanism and combined with the temporal knowledge state to predict the students’ future performance in answering exercises. Extensive experiments are conducted across four widely used KT datasets, and the experimental results demonstrate that CGKT exhibits superior performance compared to the state-of-the-art baselines.