With the rapid development of information technology, people gradually rely on social media to exchange information with the outside world. Graph neural network (GNN) has achieved excellent performance in graph node classification tasks, and it only needs to use partially annotated data to achieve good performance, making it very suitable for social network data analysis problems where it is difficult to fully annotate due to the large scale of data. Therefore, in order to deeply analyze the expression characteristics of college students on social media and the current problems and influencing factors of expression and to promptly correct the irrational communication behavior of college students’ viewpoint expression on social media, this article proposes an improved Graph U-Nets (IGUN) model for analyzing college students’ social media data. By introducing SGC and BGNN, the model can fully grasp the global and local characteristics. To verify the performance of the model, this study used pretrained model data to train the IGUN model. Through the four commonly used evaluation indicators, this article compares the effectiveness of the IGUN model with other models and verifies its good performance in analyzing social media data for college students. This provides reference value for relevant departments to carry out substantive ideological and political education work for college students in the future.