An Improved LSTM-Transformer Algorithm for University English Teaching

In recent years, with the rapid development of artificial intelligence, deep learning has helped improve the efficiency and quality of college English teaching. But the current college English teaching mode under the globalization is dependent on the disadvantages. Therefore, this paper proposes a hybrid college English teaching method, which utilizes Transformer’s powerful feature extraction and efficient sequential data processing capabilities, combined with the strong driving force and synergistic effect of target scenario model, to effectively improve teaching quality. Compared with other similar algorithms and teaching methods, it is found that the proposed algorithm tends to be stable quickly in the iterative process, and the accuracy rate is $93.6 \%$, the loss rate is $22.8 \%$, and the square root error value of the score prediction is 7.978 points. In the actual teaching simulation, the comprehensive accuracy of the proposed method in different dialogues is $\mathbf{9 1. 9 0 \%}$, and the median recognition time is only 0.17 s. These data fully prove that the method has high efficiency and accurate identification ability, which can play a good role in college English teaching with cross-cultural background, and then vigorously promote the integration of artificial intelligence and college teaching.