Currently, higher education is facing an important opportunity of digital transformation, and the traditional teaching mode of ideological and political theory class is difficult to meet the learning needs of college students in the new era. In this study, we constructed the interaction mode of college ideology and politics classroom based on dynamic data visualization, and designed a classroom behavior recognition model integrating super-resolution algorithm and 3D convolutional feature extraction. Methodologically, the Moodle platform was used to establish a blended teaching framework, and the FIAS interaction analysis system was used to encode and record the speech behaviors of teachers and students, and combined with computer vision technology to realize the automatic recognition of students’ classroom behaviors. Sixty students from University T were selected to carry out the experiment, recording 45-minute teaching videos of the Civics course and identifying five typical classroom behaviors: listening to lectures, writing, raising hands, playing cell phones and drinking water. The results show that the proposed classroom behavior recognition model outperforms the comparison model in both accuracy and mAP metrics, where the mAP reaches 0.925, which is 3.70%, 4.99%, and 2.32% higher than CycleGAN, ResNet, and YOLOv6, respectively. The FIAS analysis results show that the teacher’s verbal behaviors account for 60.78% of all the behaviors, the student’s verbal behavior accounted for 38.39%, teachers’ indirect influence accounted for 21.35%, and direct influence accounted for 39.43%. It is concluded that the dynamic data visualization technology can effectively support the innovation of the interaction mode of Civics and Political Science classroom, which provides scientific basis and technical support for improving the quality of teaching.
