Impact of Emojis on Automatic Analysis of Individual Emotion Categories

Proceedings of Recent Advances in Natural Language Processing, pages 124–131 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_014 124 Impact of Emojis on Automatic Analysis of Individual Emotion Categories Ratchakrit Arreerard, Scott Piao School of Computing and Communications Lancaster University UK {r.arreerard,s.piao}@lancaster.ac.uk Abstract Automatic emotion analysis is a highly chal- lenging task for Natural Language Processing, which has so far mainly relied on textual con- tents to determine the emotion of text. How- ever, words are not the only media that carry emotional information. In social media, peo- ple also use emojis to convey their feelings. Recently, researchers have studied emotional aspects of emojis, and use emoji information to improve the emotion detection and classifi- cation, but many issues remain to be addressed. In this study, we examine the impact of emoji embedding on emotion classification and in- tensity prediction on four individual emotion categories, including anger, fear, joy, and sad- ness, in order to investigate how emojis affect the automatic analysis of individual emotion categories and intensity. We conducted a com- parative study by testing five machine learning models with and without emoji embeddings in- volved. Our experiment demonstrates that emo- jis have varying impact on different emotion categories, and there is potential that emojis can be used to enhance emotion information processing. 1 Introduction In this study, we investigate the issue of how emojis can impact on the automatic analysis of emotion in social media messages. This topic has been stud- ied over past years, but further research is needed to fully understand the characteristics of the emo- jis and how they contribute to the conveyance of emotion. Automatic emotion analysis is a process of identifying emotions expressed by people. In social media, the emotions can be conveyed with various media including text, emojis, pictures, or other codes. Because social media platforms impose little or no restriction on language usage in terms of gram- mar and formality, social media data contains a wide range of styles and forms, including informal, colloquial, slang, and ungrammatical expressions, mixed with emojis and other images. Such an un- constrained writing styles of social media messages present a tough challenge to the task of automatic emotion processing. As Hasan et al. (2019) pointed out, the casual style and semantic ambiguity of so- cial media messages are the main two challenges in determining emotions in such data. To improve the automatic emotion analysis, researchers started to consider emojis as additional features. For ex- ample, word and emoji embedding are combined in the hope to generate better features for emotion classification. Emojis can contain emotion informa- tion that can help to identify emotions. However, as Barry et al. (2021) found, emojis are not always a good choice for representing emotion. In this work, firstly we carried out experiments of emotion classification of four emotion categories and emotion intensity prediction using word em- beddings as the sole features based on EmoInt dataset (Mohammad and Bravo-Marquez, 2017), and used the results as a benchmark. Then, we added emoji embeddings to …

Content truncated to 500 words