Sentiment Analysis of Chinese Product Reviews using Gated Recurrent Unit

Despite the explosive growth of Chinese e-commerce platforms in recent years, research focusing on the sentiment classification of Chinese documents pales in comparison to its western counterparts (English documents). This paper looks into the nascent area of Natural Language Processing (NLP) in the Sentiment Analysis of Chinese Text. The proposed Deep Learning method is the use of a sentence-based approach in the sentiment analysis of online reviews to gain more granularity and increased classification accuracy. Experimental results on a balanced (50:50), 2 class (positive, negative) test dataset of 1669 product reviews show an empirical accuracy of 87.66%, while results on an imbalanced (18:82) test dataset of 2519 product reviews show an accuracy of 87.9%, thus demonstrating the effectiveness and robustness of this proposed approach.