Intelligent technology plays a pivotal role in revolutionizing learning assessments, overcoming the constraints of traditional assessment methods and driving educational innovation. Knowledge tracing (KT) emerges as a critical component for assessing students' learning states and forecasting their future performance. However, existing graph-based KT models often ignore certain real-world factors (e.g., sequence dependencies, question information, and hierarchical relationships between knowledge concepts). These factors hinder the model's overall performance in predicting students' learning progress and identifying their knowledge deficiencies, thus limiting the model's applicability in personalized learning. To address these challenges, we propose a novel graph-based effective knowledge tracing model via subject knowledge mapping (SGKT). We begin by extracting essential features from historical learning interactions, which include the relationships between questions, the difficulty of questions, and the relationships between knowledge concepts. These knowledge concepts are seamlessly integrated with a purpose-designed subject knowledge mapping, establishing intricate hierarchical relations. Subsequently, we employ graph convolutional networks to aggregate the embeddings of questions and knowledge concepts, thereby optimizing the update process for predicting students' knowledge states. Experimental validation on the dataset demonstrates the superior performance of our model, surpassing the original KT model and other relevant approaches by up to 5% in terms of the area under the curve and accuracy. These results not only demonstrate the efficacy of our proposed model but also provide valuable insights for advancing learning assessments.