Improved text sentiment classification method based on BiGRU-Attention

Abstract

Abstract Aiming at the problem that the traditional text sentiment classification method is not sufficient for text context information learning and key feature extraction ability, this paper proposes a BiGRU-Attention based text sentiment classification method to classify Chinese texts. The method used a Bidirectional Gated Recurrent Unit (BiGRU) instead of the Bidirectional Long Short-Term Memory network (BiLSTM) to build a hidden layer, and introduces an attention model to input the result of each time point in the hidden layer to the fully connected layer yields a probability vector. Then use this probability vector to weight each hidden layer result and add it to get the result vector. The experimental shows that the model this paper proposed has better accuracy and effectiveness in text classification.