The wave of education informatization sweeps in, providing new opportunities and challenges for cultivating new talents. This study applies deep learning and learning theories related to combinatorial mathematics to college English teaching in order to improve the teaching effect, and builds a smart classroom teaching model of college English from three dimensions: guiding, advancing and strengthening. Using quantitative big data analysis methods, cluster analysis and correlation analysis of students’ learning behaviors are conducted to explore the group portrait of students in English courses. Then a teaching comparison experiment is conducted to verify and evaluate the application effect of the English smart classroom teaching model. The sample students were clustered into four types: excellent, diligent, maintenance and self-abandonment, with the diligent and maintenance types accounting for the highest proportion of students, 31.39% and 45.26%. The students’ English test scores and innovation cultivation effects as a whole improved by 5.65% and 16.24% respectively after the teaching experiment. The results show that the teaching design based on deep learning can effectively improve the teaching effect of college English smart classroom and promote the development of students’ innovative thinking and ability.