Emojis-Based Recurrent Neural Network For Chinese Microblogs Sentiment Analysis

As a novel graphical expression derived from emoticons, emojis have a wide application in social networks. Emojis can assist people in expressing stronger sentiment or show subtler sentiment indirectly which are helpful to sentiment analysis. In this paper, we propose an emojis-based recurrent neural network for sentiment analysis in Chinese microblogs. Firstly, we differentiate ambiguous emojis and explicit emojis by using pre-trained word embedding and a new sentiment lexicon. Then we verify that users' information can eliminate the ambiguity of ambiguous emojis to some extent and confirm the sentiment polarity of ambiguous emojis. On the basis, we obtain emoji representations by utilizing the position vector, semantic vector and sentiment vector of emojis, then put the emoji representations into Bi-directional gated recurrent unit(BiGRU) neural network model to conduct sentiment analysis. The experimental results on a Chinese microblog dataset demonstrate that compared with other baselines, the proposed model can improve the accuracy significantly in sentiment analysis.