Sentiment Analysis Augmented by Emoticons

Abstract

Social media platforms are the main resources to collect people's sentiments and opinions. We can extract quantities of useful information from the social network. Weibo is the most popular social networking application in China. In this paper, we'll describe our attempts at producing a state-of-art Weibo sentiment classifier using CNN, LSTM and existence of emoticons in users' microblogs. The experiments carried out on standard datasets including 120,000 microblogs and then group them into positive and negative sentiments. The models include character-based classifier and emoticon-based classifier. To boost performances, we assembled character-based classifier and emoticon-based classifier together to realize a compound classifier. We also implemented necessary experiments to measure the accuracy. The final results prove that emoticons in microblogs can improve the performance of traditional sentiment classifiers.