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

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