Cognitive model is playing very important role in predicting students’ performance and recommending learning resources. Thus, it has received a great deal of attention from researchers. However, most of the existing work design models from the aspect of students, ignoring the internal relation between problems and skills. To address this problem, we propose a deep cognitive diagnosis framework to obtain students’ mastery of skills and problems by enhancing traditional cognitive diagnosis methods with deep learning. First, we model the skill proficiency of students according to their responses to objective and subjective problems. Second, students’ mastery on problems is modeled based on attention mechanism and neural network, considering both the importance and the interactions of skills. Finally, considering the facts that students may carelessly select or simply guess the answer, we predict students’ performance via the proposed model. Extensive experiments are carried out on two real-world data sets, and the results have proved the effectiveness and interpretability of this work. •We model the skill proficiency with IRT.•The importance of the skills is acquired based on attention mechanism.•The skill interaction is obtained and quantified by a neural network.