Sentiment Analysis Based On Deep Residual Bidirectional Gated Recurrent Unit Neural Networks

Abstract

Human beings have rich emotions, in which positive emotions need to be constantly maintained, while negative emotions need to be regulated since most mental diseases are caused by the long-term persistence of negative emotions. Aiming at emotion classification in sentiment analysis tasks, a deep residual BiGRU (Bidirectional Gated Recurrent Unit) neural network model is introduced to improve the classification effect and solve the problem of long-distance dependence between layers of deep RNN (Recurrent Neural Network) by using a recurrent correlation channel connects all layers and a loss function with CRF (Conditional Random Fields) technique. This model provides a preferable method and outperforms other commonly used models in extracting sentiment features and recognizing emotions expressed in texts. Experiments on simplified Chinese sentiment analysis data set show that it classifies and predicts emotions more accurately than stacked neural network models built on RNN, TextCNN (Text Convolutional Neural Network), BiLSTM (Bidirectional Long Short-Term Memory), or non-neural network models built on NB (Naïve Bayesian) classifier and LR (Logistic Regression) method.