Abstract: In order to combine forgetting features with knowledge states and comprehensively analyze their joint effects on predicted answers, a knowledge tracking method based on forgetting gated deep attention memory is proposed. The weight optimization forgetting gating mechanism was incorporated into the attention memory structure, and the potential conceptual information proportion was adjusted by optimizing the weight to optimize the performance of information capture ability. Based on the constantly developing knowledge state of students, we captured embedded representations of potential concepts and their relationships from the dynamic potential concept map, and used their useful information to sort the problems. Experimental verification was conducted on four datasets, and the results demonstrated the superiority of the proposed method.