In this paper, we propose a learning input assessment model based on multi-source data fusion, which analyzes students’ expression features with VGG16 network, acquires mouse track data corresponding to students’ expressions based on deep neural network, and then deeply fuses students’ log data and interaction data based on the data conversion algorithm in the event window. The model is utilized to establish a smart classroom, based on which the multidimensional learning characteristics of 50 students are studied, thereby promoting the application of multi-source data fusion technology in the education sector. The multi-source data fusion model can effectively recognize the emotional changes of students in the smart classroom, and the accuracy of multi-emotion recognition is more than 95%. In the smart classroom, the students’ mouse browsing trajectories focused on the web pages of “Teaching and Learning” and “Teaching Tools and Resources”, with a total click frequency of 1959 and 2005, respectively. Based on the above characteristics, the model classified the 50 students into five categories: focused students, persistent students, persistent students, occasional students and abandoned students, and the average learning input values of each category of students were 4.55, 3.57, 2.48, 1.48, and 0.54, respectively. The construction of the smart classroom is helpful for the wide dissemination of the model.
