Dual-Channel CNN-BiGRU Sentiment Analysis Method Based on Part of Speech
Download as PDF
DOI: 10.25236/iiiece.2022.029
Corresponding Author
Xin XU
Abstract
For less research on lexical to emotional tendency in sentiment analysis, traditional CNN model sentiment classification method will have classification inaccuracy problem, this paper proposes a dual-channel CNN-BiGRU sentiment analysis method research method based on part of speech, can use rules to extract the corresponding lexical text combined with the original text to construct a dual-channel input to the neural network, the CNN can extract local semantic features, the Bi-GRU is used to extract global features containing context, which are fused with local features to complement them, and finally the fused features are input to the classifier for sentiment tendency determination. The experimental results on Chinese dataset show that the proposed-to-model approach in this paper has improved in accuracy, recall, and F-value compared to traditional neural networks.
Keywords
Part of speech, Cnn, Bigru, Dual channel