Sentiment Analysis of Comment Texts Based on CNN-BiGRU-Attention

Abstract

In recent years, deep neural network models have become more and more widely used in the field of sentiment analysis. Compared with traditional machine learning algorithm-s, Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Convolutional Recurrent Neural Networks (CRNN), etc are more stable and accurate. However, these algorithms are prone to overfitting during the experiment, resulting in lower accuracy and precision. Therefore, in this paper, we propose a feature extraction method based on convolutional neural network (CNN) and bidirectional GRU model, the two features are fused in the full connection layer, which contains both the text features and the semantic relationship between the front and back of the text, then we output the results through a layer of attention mechanism and softmax function. Finally, after many experiments, it has been proved that our model works better than baselines in the accuracy, precision, recall rate, and F value by 3 to 5 percentage points, which proves the feasibility of the proposed model.