Chinese Microblog Sentiment Analysis by Adding Emoticons to Attention-Based CNN

Abstract

Nowadays, people are used to sharing their views and ideas on social media platforms, which generates enormous amounts of data every day. This research adopted sentiment analysis to disclose embedded information in Chinese short texts, which can serve as an integral part of social media monitoring and analytics. The research proposed a deep learning method, Attention-of-Emoticons Based Convolutional Neural Network (AEB-CNN), by integrating emoticons and attention-based mechanisms with CNN to enhance the accuracy of sentiment analysis. An implementation was carried out by TensorFlow; the accuracy of sentiment polarity of Chinese microblogs reached somewhere between 85.1% and 89.1% while achieving shorter execution time compared to other methods when the size of training dataset ranged from 10,000 to 30,000 sentences.