Sentiment Analysis of Hybrid Network Model Based on Attention

Abstract

The existing text sentiment analysis models based on deep learning and neural network usually have problems such as incomplete text feature extraction and failure to consider the impact of key information on text sentiment tendency. Based on the parallel hybrid network and the two-way attention mechanism, an improved text sentiment analysis model is proposed. The model first takes the word vector trained by the BERT language model as the input, and then extracts the global and local features of the context simultaneously through the parallel hybrid neural network constructed by the Convolution Neural Network (CNN) and The Bidirectional Gated Recurrent Unit (BiGRU), so as to improve the feature extraction ability of the model. It also integrates the dual-way attention mechanism to strengthen the key information in the global feature and local feature, and the feature vectors obtained by feature fusion are used for sentiment analysis.