Sentiment Classification with Gated CNN for Customer Reviews

Recurrent neural networks(RNNs) have been applied to sentiment classification but RNNs is usually heavier than convolutional neural networks (CNNs), turning more interest in the application of CNNs to language tasks. In this paper we propose a method to apply gated CNN (gCNN) with Maxpooling to sentiment classification of customer reviews. In our proposal, the application of gCNN is to sentiment classification, instead of constructing a language model. Our experiment is conducted with Amazon Product Review dataset and Japanese review dataset of TripAdvisor. The whole of each review is used as an input, instead of each sentence. The result is that a simple application of gCNN to sentiment classification achieved sufficient accuracies with the two datasets. Thus, an implication is that gCNN is proven to work fine for sentiment classification much faster than RNNs with fine results in the different language datasets.