A Deep Learning Model Enhanced with Emojis for Sina-Microblog Sentiment Analysis

In Sina-microblog comments, abundant emojis, as an important piece of information for the sentiment analysis, are used to vividly express users's opinions and feelings. To classify timely and effectively the sentiment hidden in the vast pool of the microblog comments, this paper proposes an emojis-enhanced sentiment analysis model. The model first extracts emojis from review sentences and constructs emoji sequences as an enhanced information, and then uses BiLSTMs to learn text-based representation. Furthermore, this model adopts a new attention mechanism to incorporate contextual information of sentences into emojis to learn the emoji-based auxiliary representation. Finally, the final representation is obtained by combining text representation and emoji representation through the full connection layer. Extensive experiments show the effectiveness of the proposed method on sentiment classification for Chinese microblog comments.