A Hybrid Neural Network Optimal Model Based on Multi-Channel CNN and BiLSTM with Attention Mechanism

In order to figure out the issue that the current CNN and RNN do not extract enough text features and fail to capture key information when performing sentiment analysis tasks, this paper proposes a hybrid neural network model WCAFB with attention mechanism, which combines CNN and BiLSTM. The proposed model uses two pre-trained word vector models, Word2vec and Fasttext, to learn word vectors in combination with embedded layer fine-tuning. The proposed model uses multiple CNN with different sizes of convolutional kernels to capture local features of text at different scales and integrates attention mechanisms into the extracted features to dynamically adjust their weights. BiLSTM is used to extract semantic information from the context, and then the local and global features extracted from CNN and BiLSTM are fused. Finally, use the Softmax classifier to obtain the sentimental polarity of the text. The experimental results indicate that this model can extract more abundant text features than other single models, and the classification accuracy has been improved to varying degrees compared with other models.