Impact of Emojis in Emotion Analysis on Code-Mixed Text

With the maturity of sentiment analysis technology, researchers are gradually not satisfied with only excavating positive and negative attitudes from the text but hope to obtain more delicate emotions. In addition, the development of social media makes people from different backgrounds like to express their emotions and feelings on social platforms. However, textual communication from social media is too colloquial, leading to the code-mixed text phenomenon, where sentences contain different languages, which poses difficulties for text analysis research. We observed that emojis in text contain emotion content and are universal across various language texts. This paper proposes a deep learning method for multi-class code-mixed emotion analysis using emojis. The proposed method has achieved good results on media text datasets and attempts to solve the difficulty of mixed sentiment classification with different codes. In this work, experiments were carried out on Hindi-English and Chinese-English code-mixed datasets, and the results have shown that integrating emoji representation with multilingual language model yielded 4% improvement on emotion analysis. Emojis should be preserved in emotion analysis regardless of which language the text is in.