An integrated non-linear deep learning method for sentiment classification of online reviews

This paper presents an integrated non-linear method for sentiment classification of online reviews. Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNNs) utilizing Bidirectional LSTM (BiLSTM) are adopted in the research as they have demonstrated great information displaying abilities when testing large datasets from a wide scope of application regions. CNNs offer points of interest in choosing good level features and LSTM have demonstrated great capacities of learning sequential (non-linear) information score encoded via BiLSTM. Random Forest is then utilized to perform the learning process for large sample of 23K and small sample size of 11K tweets. The results show that the proposed integrated non-linear deep learning method has a better accuracy as compared to other existing methods.