Pattern Recognition and Deep Semantic Network Analysis Techniques for Rhetorical Devices in English Literary Texts

English literary discourse contains rich aesthetic elements and rhetorical devices, and its teaching is very suitable for penetrating aesthetic education. In this study, we constructed a bidirectional multi-angle matching model architecture based on convolutional neural network to realize the recognition of rhetorical devices in English literary texts, and designed a deep semantic network based on LSTM to analyze the rhetorical emotion expression in English literary texts. The experimental validation and comparison results show that using the LSTM-based complex network text sentiment analysis model proposed in this paper achieves good results in various types of metaphorical sentiment analysis, and this paper’s model improves 1.9% in the F1(avg.) metrics compared to the TRAT-LSTM model. In the LSTM weight analysis of a sample sentence, it is found that this paper’s method can effectively find that there are small differences in sentence weights and the sum of word weights is much less than 1, which leads to the effectiveness of this paper’s method.