Emoji-based Fine-Grained Attention Network for Sentiment Analysis in the Microblog Comments

Microblogs have become a social platform for people to express their emotions in real-time, and it is a trend to analyze user emotional tendencies from the information on Microblogs. The dynamic features of emojis can affect the sentiment polarity of microblog texts. Since existing models seldom consider the diversity of emoji sentiment polarity, the paper propose a microblog sentiment classification model based on ALBERT-FAET. We obtain text embedding via ALBERT pretraining model and learn the inter-emoji embedding with an attention-based LSTM network. In addition, a fine-grained attention mechanism is proposed to capture the word-level interactions between plain text and emoji. Finally, we concatenate these features and feed them into a CNN classifier to predict the sentiment labels of the microblogs. To verify the effectiveness of the model and the fine-grained attention network, we conduct comparison experiments and ablation experiments. The comparison experiments show that the model outperforms previous methods in three evaluation indicators (accuracy, precision, and recall) and the model can significantly improve sentiment classification. The ablation experiments show that compared with ALBERT-AET, the proposed model ALBERT-FAET is better in the metrics, indicating that the fine-grained attention network can understand the diversified information of emoticons.