Smart classrooms have become a significant trend due to the high pace of development and extensive use of artificial intelligence and digital technologies, which opens up new opportunities to innovate the assessment of Civics and Political Science education. The current paper presents MASA theory as seen through the eyes of the meta universe and constructs a multimodal classroom information acquisition framework with the goal of facilitating paradigm shift and functional improvement in the process of course evaluation. Specifically, a better EfficientNet-driven model is created to recognize the facial expressions in class, then a more effective DenseNet-driven model is trained to recognize the learning behavior and action of students, which are both evaluated on a self-made dataset. Based on this, an intelligent evaluation method of Civics teaching is built based on the estimation of probability and the combination of learners expression with behavioral performance. The practical application outcomes show that out of ten students, Students 1, 2, and 3 have the best learning condition with each recording a score of over 0.400 and the other ones require further improvement. The offered approach will be able to track the status of the students during the learning process, and successfully complete the evaluation procedures of the Civics course. Such results enable offering more specific guidance and counseling, which in turn supports the overall, healthy, and sustainable development of students.
