A Sentiment Analysis Method Based on Bidirectional Long Short-Term Memory Networks

Abstract Although the traditional recurrent neural network (RNN) model can cover the time information of the whole sentence theoretically, the gradient is dominated by the short-term gradient, and the long-term gradient is very small, which makes it difficult for the model to learn the long-distance information, and thus the effect of RNN on long text sentence recognition is poor. The long short-term memory network (LSTM) introduces the gate mechanism, especially the forgetting gate, which improves the disappearance of the gradient of RNN. Therefore, LSTM can store long text information and remove or increase the ability of information interaction by adding gate structure, which has natural advantages for long text processing. Based on the word vector matrix of GloVe model, on the open-source comment sentiment140 data set, we use the TensorFlow framework to construct the LSTM neural network and divide the data into the training set and test set based on the ratio of 4:1, design and implement the sentiment analysis published by Twitter users based on LSTM model, and then propose the bidirectional LSTM (Bi-LSTM) sentiment analysis method. The experimental results show that the accuracy of bidirectional LSTM is higher than that of unidirectional LSTM in sentiment analysis.