Recognizing Metaphorical Expressions in Chinese Speech and Their Natural Language Processing Strategies

Metaphor, as a special phenomenon in natural language, is of great significance for natural language processing tasks such as sentiment analysis, machine translation, and question and answer systems. In this paper, we design a model for metaphor recognition based on grammatical structure and word meaning in Chinese speech. The model combines several key techniques of metaphor recognition, such as using the TP-IDF algorithm as the feature extraction model of Chinese speech text, and using Bi-LSTM as the central network structure of the metaphor recognition model. Finally, the performance of this paper’s model in recognizing metaphors in Chinese speech is analyzed through an experimental design. When the number of layers of attention mechanism in this model is 4, Precision, Recall, and F1 are 94.32%, 95.03%, and 93.36% respectively, and the metaphor recognition effect is optimal. The TF-IDF feature extraction algorithm adapts well to the metaphor recognition model constructed in this paper. The model of this paper has good recognition effect on five types of emotions except “surprise”, and the F-value of MI_SS+MI_WS model for metaphor recognition in Chinese speech is improved by 12.92%~26.18% compared with the comparison method. This study promotes the development of metaphor recognition techniques in Chinese speech and provides new perspectives and strategies for other tasks in natural language processing.