Accurately diagnosing students’ mastery of knowledge concept (KC) is the premise of personalization learning resources recommendation and teaching students in accordance with their aptitude. The classic cognitive diagnosis model assumes that knowledge con-cepts are isolated. In fact, there is a dependency between KCs, especially when recom-mending personalized learning resources based on cognitive diagnosis results, the depend-ency between KCs will seriously affect the recommendation performance. Based on this, in this paper, a personalization exercise recommendation method based on graph cognitive diagnosis is proposed. This method includes graph-based cognitive diagnosis and graph-based personalized exercise recommendation. This method introduces predecessor knowledge concept and successor knowledge concept in concept map and the importance of knowledge concept into the Item Response Theory (IRT) model to diagnose the knowledge mastery of each student. According to the diagnosis results and the students’ current ability level, select several KCs that are most important to the student, and then select exercises related to these KCs from the exercise bank and recommend them to the student to learn. Extensive experimental comparisons show that our proposed method is in line with the idea of constructivist learning theory, and can recommend exercises for stu-dents who have not mastered knowledge concepts well. It has good accuracy and more intuitive when recommend test exercises.