Graph neural networks are widely used in educational research, and have strong application potential in the prediction of students’ comprehensive development and recommendation of personalized educational resources. In this paper, the information and characteristics of students are mined from massive learning data, and the prediction method of multi-topology graph neural network is used to realize the effective prediction of students’ comprehensive development. Through the graph neural network, knowledge graph and cluster search algorithm and other technologies, the personalized learning path planning and optimization are completed, and the personalized learning path is designed. The research shows that the data accuracy of the student development trend prediction model in this paper reaches the qualifying value of 0.1, and the absolute maximum value of the error does not exceed 0.17, so the model constructed in this thesis is effective and robust. It can fulfill the task of student development direction prediction. The usage frequency of generating learning paths are more than 60%, so the learning path generation method proposed in this paper is practical. And the average grade of the users who use this method is 6.17 points higher than the average grade of the users who do not use this method.