Application of CNN-BiGRU Model in Chinese Short Text Sentiment Analysis

In view of the fact that traditional sentiment analysis[1] techniques require manual annotation and construction of sentiment lexicon, and the current use of deep learning techniques for sentiment analysis, there is a need for manual screening and large workload, and the accuracy and training speed need to be improved. This paper proposes a text sentiment analysis model (CNN-BiGRU) combining a convolutional neural network (CNN) and bidirectional gated recurrent unit (GRU). The model first uses CNN to extract the local static features of the text, and then extracts the sequence features of the text through BiGRU, and then the two unidirectional GRU layers further reduce the extracted features. Finally, the Sigmoid classifier is used for the final sentiment classification. Through the self-built Douban film and television commentary dataset, compared with the CNN-BLSTM model, the classification accuracy rate increased by 2.52%, and the training rate increased by 41.43%, which verified the validity of the model.