In recent years, the psychological problems of college students could not be ignored, as they have seriously affected the growth of students and the normal teaching order of colleges and universities. However, there exists a strong noise in college students' psychological sample data set and a strong correlation between its data. Aiming to solve this problem, this paper proposes a psychological crisis warning method for college students based on big data mining and structural equation model (SEM). This method is oriented to massive user data in social networks. Particle swarm optimization is introduced to improve the random forest algorithm, and the original data is labeled to alleviate the impact of data noise on the recognition accuracy. The simulation example comes from an efficient actual data set in the southern China. The experimental results show that the proposed method can achieve an efficient analysis of actual complex data, and can provide reliable psychological auxiliary diagnosis for practitioners.