The development of social media has brought many tests to the mental health education of college students, and some college students have fallen into network addiction and dependence, which greatly affects their physical and mental health. The article uses microblogging social media as the source of students’ mental health data and preprocesses the data using data de-emphasis and Chinese word separation. It also analyzes the problematic manifestations of students’ mental health in colleges and universities, extracts students’ mental health indicators by using the TF-IDF algorithm, and realizes the recognition of students’ mental health topics by using the BTM model. The CNN-LSTM-ATT model was established by introducing the attention mechanism and LSTM model to assess the mental health status of college students. The data was analyzed in terms of students’ mental health characteristics predictive validation and used to develop intervention strategies for students’ mental health. The text length of students’ mental health is [1,22], which occupies 86.98% of all sentences, and the AUC value corresponding to the BTM model is 0.946, and the prediction accuracy of the CNN-LSTMATT model for the assessment of students’ mental health in colleges and universities can reach up to 97.62%. The social media data can clarify the mental health status of college students and realize the construction of students’ mental health intervention strategies from the dimensions of students’ media literacy and regulatory mechanisms.