EmoTag – Towards an Emotion-Based Analysis of Emojis

Proceedings of Recent Advances in Natural Language Processing, pages 1094–1103, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_126 1094 EmoTag – Towards an Emotion-Based Analysis of Emojis Abu Awal Md Shoeb Shahab Raji Gerard de Melo Department of Computer Science Rutgers University, New Brunswick, NJ, USA {abu.shoeb, shahab.raji, gerard.demelo}@rutgers.edu Abstract Despite being a fairly recent phenomenon, emojis have quickly become ubiquitous. Besides their extensive use in social me- dia, they are now also invoked in customer surveys and feedback forms. Hence, there is a need for techniques to understand their sentiment and emotion. In this work, we provide a method to quantify the emo- tional association of basic emotions such as anger, fear, joy, and sadness for a set of emojis. We collect and process a unique corpus of 20 million emoji-centric tweets, such that we can capture rich emoji seman- tics using a comparably small dataset. We evaluate the induced emotion profiles of emojis with regard to their ability to pre- dict word affect intensities as well as sen- timent scores. 1 Introduction In recent years, information technology has pro- foundly altered the way humans communicate. A substantial proportion of the global population has adopted the use of social media platforms (such as Twitter, Facebook, and Instagram) and mes- saging technology (such as Facebook Messenger, WeChat, and WhatsApp) to interact and voice their opinion. The unique properties and expres- sive capabilities afforded by computer and mobile device-mediated communication has led to quite distinct forms of expression in comparison with classic email etiquette, let alone traditional writ- ten correspondence. Meanwhile, for any meaningful analysis of so- cial interactions or expression of opinions, it is critical to extract and understand the sentiment and the affect of the source. There are numer- ous studies investigating the connection between words or sentences and the affects they convey. However, emojis are a particularly prominent fea- ture of modern online interaction. Thus, this pa- per introduces a new basis for studying this new modality with regard to conveyed affective associ- ations. Emojis have become widespread in social media, and are variously used to carry emotional and contextual information pertaining to the con- tent of social media posts. There have been studies exploring the relationship between hashtags and tweets (Ferragina et al., 2015), and between emo- jis and tweets (Campero et al., 2017). Additional research has aimed at conducting sentiment anal- ysis based on emojis and hashtags (Novak et al., 2015). A number of other works study the con- nection between words and emotions, resulting in datasets such as EmoLex (Mohammad and Turney, 2013). Most of these studies relied upon a crowd- sourcing approach to compile the data and lexi- cons and to capture relationships among linguistic and paralinguistic elements (Kulahcioglu and de Melo, 2019). However, previous work has neglected to focus on the emotional aspects of emojis. For instance, we may ultimately be interested in devising a sys- tem that jointly assesses the affect conveyed by a tweet based not only on the words, but also in …

Content truncated to 500 words