Based on the traditional Flanders interaction analysis method, this paper proposes a new method for analyzing English teachers’ behavior by combining long and short-term memory networks in deep learning techniques. Firstly, the Flanders coding system is optimized in the coding stage, and the English teacher behavior categories are refined to better reflect the integrity, richness, and diversity of the classroom teaching process so as to construct an English teacher behavior dataset. Then, we designed the research protocol, determined the research subjects and instruments, classified the time series representing English teacher behavior based on the LSTM English teacher behavior analysis model in the results and analysis phase, and analyzed and evaluated the correlation between teacher behavior of students’ English learning achievement based on the classification results. The results show that the probability of getting the English teacher behavior evaluation index categories 1 to 4 is 0.6111, 0.00049, 0.002, and 0.3882, respectively. There is a significant correlation between both student learning achievement and teacher behavior, and the correlation coefficient of both is (r =0.73). Teacher behavior affects the achievement of student learning effectiveness. This study helps teachers to reflect on their teaching behaviors so as to improve the quality of teaching in higher education.