An Embedding Method for Sentiment Classification across Multiple Languages

Abstract

Embedding methods are feature representations of words, which are able to capture both semantic and syntactic information from contexts. However, existing embedding methods for learning context are typically unable to capturing sufficient sentiment information. In this study, we conduct an investigation on how to improve the performance of sentiment classification using sentiment embedding approach. Particularly, we first present a new word embedding method based on a supervised method to capture the semantic sentiment information. Then, Deep Learning models, by combining different architecture such as Recurrent Neural Network (RNN), Convolutional Neural Network (CNN), and attention mechanisms are developed for improving the performance of the sentiment classification. The evaluation on well-known bench-mark datasets with different languages (i.e., English and Vietnamese) indicates the promising results of our method.