Knowledge tracing models the learner's state of each knowledge point, infers overall knowledge state, and predicts their future learning performance. However, existing research does not incorporate the relationships between knowledge point states when modeling the overall knowledge state, which affects the final prediction accuracy. To address this problem, this paper proposed a knowledge tracing model that integrated the relationships between knowledge point states. Firstly, it vectorized the knowledge points in the dataset, and constructed a knowledge point representation graph. Secondly, it diffused the knowledge point representation graph to the latent space to reflect its inherent structure and essential information. Thirdly, it fused the current exercises and knowledge points as guiding vectors to extract the knowledge point state graph from the latent representation of the knowledge point representation graph. Fourthly, based on the knowledge point state graph, it derived the overall knowledge state. Through experiments comparing four related models on three datasets, the proposed model demonstrated certain advantages in AUC, ACC, and DOA. Among them, it performed best on the ASSISTments2009 data set. Compared with the optimal value and the lowest value in the comparison model, AUC increased by 1.17% and 10.57% respectively, and ACC increased by 3.23% and 12.17% respectively, indicates quality increased by 1.95% and 10.40% respectively. Furthermore, the internal reason process of the knowledge point state and its relationships are visualized, as well as their correspondence with the actual test results, demonstrating the model's interpretability. Additionally, when applied to predicting student performances in three courses and compared with related models, the proposed model achieves better results, demonstrating its practicality.