CNN blend ERNIE for sentiment classification
YunPeng Xu, Hui Cao · 2023
For Chinese text sentiment classification task, this paper propose the CNN blend ERNIE for Sentiment Classification (CSEC) model, it concatenates all CLS vectors from twelve Encode Layers of Enhanced Representation through Knowledge Integration (ERNIE) by introducing Attention Mechanism (AM) then sends the combined vector to Convolutional Neural Networks (CNN). It aims to learn different levels of information from different layers. Through different experiments to validate, CESC achieves the best results.