Knowledge Tracing (KT) is a technique in the field of education, primarily tasked with predicting students’ mastery of a particular knowledge concept by analyzing their learning history data. Although recent research has employed various deep neural network architectures to address the KT problem, existing KT methods have not effectively utilized the difficulty-ability interaction effect, which is highly relevant for assessing a student’s knowledge state. In this paper, we focus on the effect of the difficulty-ability interaction on knowledge state assessment and propose the Difficulty-Ability Matching Knowledge Tracing (DAMKT) model. Specifically, we first introduce a difficulty encoder and an ability encoder to obtain question difficulty embedding and student ability embedding, respectively. Then, to establish the relationship between a student’s knowledge state, question difficulty and student ability during the learning process, we design an adaptive neural network that is divided into three modules: (1) personal difficulty perception; (2) personalized knowledge gain; and (3) differential knowledge forgetting. Finally, we conducted extensive experiments on four real-world datasets, and the results indicate that DAMKT outperforms other baseline models.