Research on the Dynamic Prediction Model of Student Behavior in the Information Management System of Ideological and Political Education for Student Groups in Colleges and Universities

Abstract

This study proposes a dynamic prediction model of student behavior based on multilayer fragment dynamic semantic spatio-temporal graph convolutional network (MF-STGCN) to meet the needs of information management of ideological and political education in colleges and universities. By integrating heterogeneous data from multiple sources, such as campus card consumption, WiFi track, course schedule, etc., a three-level data system of “activity-behavior-abnormal behavior” is constructed. First, activities are defined and clustered into five types of functional areas (teaching, dormitory, cafeteria, etc.). Secondly, we extracted the spatial and temporal characteristics, calculated the proportion of students' stay time in each area on a weekly basis, and formed the spatial and temporal vectors to characterize behavioral preferences. Finally, the MF-STGCN model is designed, which innovatively introduces a multi-fragment segmentation mechanism, combining dynamic semantic graph convolution, temporal convolution and fully connected layers to realize multi-scale feature fusion. The empirical study shows that the entropy value of WiFi usage in the teaching area is the lowest (0.18-0.41), reflecting the class schedule constraints. The dormitory has medium entropy value (0.47-0.72), which rises to 0.68 at the end of the period (schedule disruption). The cafeteria has the highest entropy value (0.65-0.82), which is consistent with fragmentation. MF-STGCN has an average relative error of only 4.23%-5.65% at the size of 100-10,000 people, which is 2.93% lower than K-nearest neighbor and GCN -4.53%, and the operational efficiency is improved by 5-9 times, with 10,000-person prediction taking 35.53 seconds. The model effectively improves the accuracy and timeliness of behavioral prediction, and promotes the transformation of civic education management from empirical decision-making to data-driven.