Japanese Sentiment Classification Using Bidirectional Long Short-Term Memory Recurrent Neural Network
Conventional sentiment classification techniques often require polarity dictionaries to train the classification model. Those approaches however, require highclabor cost for the dictionary creation process. Given the current neural network learning advancements, we try to generalize and simplify these labors. We propose a sentiment classification model based on the Bidirectional Long Short-Term Memory (BiLSTM) network over the distributed word representation. Investigating the effectiveness of feature and dictionaries, we also append the network hidden layer with the Part of Speech tag (POStag) feature and Japanese polarity dictionary information. During our preliminary experiments, we found interesting relationship between those given features. These findings lead our model to achieve the state-of-the-art performance in Japanese sentiment classification task.