Online learning models are rapidly emerging. By tracking the relevant learning behaviors of students in the learning process, certain suggestions can be provided for learners’ study guidance. In this paper, the improved K-Means algorithm and behavioral data from students’ online learning are used to cluster their learning status. In addition, the self-attention mechanism is introduced on the basis of TCN, and the network structure for dealing with the language habit classification problem is constructed as a whole by combining it with the Meier filter bank alternative network. Finally, TCNSA is used to learn and classify pronunciation habits, and the two networks are trained together to form an end-to-end recognition method. The classification effectiveness of the method was finally verified on different datasets, and the classification accuracy of Region 5 language habits increased from 42 to 49, and the classification accuracy of the remaining regions also increased compared to that of the TCN. It comprehensively reflects the performance of the TCNSA model in distinguishing language habits. The method proposed in this paper helps to provide targeted online business English learning guidance to students from different regions, with different learning behaviors, and different language habits.