Classifying Wu-Qing Couplets and General Couplets with Structural and Semantic Features

The couplet custom is one of China’s intangible cultural heritage, and it occupies a pivotal position in Chinese traditional culture. The general couplet requires that the first sentence and the second sentence be opposite and semantically coherent. As a particular type of couplets, Wu-Qing couplets are related on the character level but irrelevant at the sentence level. Such characteristics make the Wu-Qing couplet significantly different in structure and semantics compared to the general couplets. In this paper, Convolutional Neural Network is used to extract structural features. Long Short-Term Memory network and vector operations are used to extract semantic features, and the Attention mechanism is used to strengthen structural and semantic information. Finally, we propose a model with structural and semantic features between general couplets and Wu-Qing couplets. The final F1 score reached 82.6, which was an increase of 5.4 compared to the baseline model.