Methods and Implementation Paths for Evaluating the Impact of Deep Learning Based on Big Data on Ideological and Political Education and Mental Health of College Students

In the process of evaluating the learning effect of college students’ civic politics and mental health courses, there is a strong nonlinear relationship between the data involved, which leads to the problems of poor data fitting ability, low accuracy and stability of learning effect evaluation, so we propose a method for evaluating the in-depth learning effect of college students’ civic politics and mental health courses based on big data. Based on the online learning platform and LMS learning management system, students’ learning behavior data are extracted; the radial basis function (RBF) kernel function is used to fill in the missing values of the extracted data and extract the data feature values; the hierarchical analysis method is used to comprehensively analyze the learning behavior data and the corresponding feature information, construct the evaluation system of in-depth learning effect, and obtain the evaluation indexes of in-depth learning effect of Civic and Political Science and Mental Health Courses; using the 9-point comparison scale and expert scoring method, we assign scores to the degree of importance of the evaluation index data, calculate the corresponding weight coefficients of the indexes, and reduce the impact of data nonlinearity on the final evaluation results; based on the big data method, we introduce momentum on the basis of BP neural network topology, and apply the improved BP neural network algorithm to perform the steps of initialization of the network weights and learning parameters, forward propagation of the inputs, backward propagation of the error, and iterative iterations to realize the evaluation of college students’ civic politics and mental health. Steps to realize the evaluation of in-depth learning effect of college students’ civic politics and mental health courses are provided. The experimental results show that using the design method for deep learning effect evaluation, the data prediction accuracy is high, and the maximum difference is no more than [Formula: see text]. Moreover, under the situation of different data volume, the output model network training value and the actual students’ deep learning effect evaluation value have a high degree of fit, and the highest value of positive residual fluctuation is only [Formula: see text], and the negative residual fluctuation is −0.18, with the value of G being [Formula: see text], and the value of F1 being [Formula: see text], which has a good performance of [Formula: see text], and the value of F1 being [Formula: see text]. The value is [Formula: see text], with good robustness (stability and reliability) and comprehensive evaluation effect.