Emotion classification method and system based on word-phrase attention mechanism

The invention provides an emotion classification system and method based on a word-phrase attention mechanism, and the method proposes a shallow feature extraction model based on a word attention mechanism and a deep feature extraction model based on a phrase attention mechanism on the basis of a TCN. Auxiliary information contained in words, phrases and overall comments and different contribution degrees can be effectively mined, and more accurate sentiment classification performance is achieved with lower computing resources. Experiments show that the performance of the SC-WPAtt method provided by the invention is superior to that of a traditional method. 本发明提供一种基于"单词-短语"注意力机制的情感分类系统和方法，所述方法在TCN的基础上，提出的基于单词注意力机制的浅层特征提取模型和基于短语注意力机制的深层提取模型，能够有效挖掘单词、短语和整体评论所蕴含的辅助信息以及不同贡献程度，以更低的计算资源实现了更精准的情感分类性能。实验表明本文所提的SC-WPAtt方法性能优于传统方法。