Knowledge tracking makes predictions regarding students’ future learning performance by leveraging their past answer records. These predictions rely on both the historical answer data and the connections between knowledge concepts. In deep learning, its mini-batch input and end-to-end learning method risks ignoring global information like exercise difficulty, student status, and exercise correlation. Moreover, the form of input data makes predictions uninterpretable. To address these issues, we developed a framework to add crafted global information within mini-batches for the network to learn integrated insights. We designed exercise difficulty index, student confidence index, and exercise relationships as global information. Experimental findings demonstrate that the network enhanced with the added global information can markedly improve the performance of the model. On the knowledge tracing dataset, merely by incorporating global information into each batch, we achieved an average increase of over 3.2% in the AUC metric. Simultaneously, the designed artificially global information also boosts the model’s interpretability.