How Gender and Skin Tone Modifiers Affect Emoji Semantics in Twitter

Proceedings of the 7th Joint Conference on Lexical and Computational Semantics (*SEM), pages 101–106 New Orleans, June 5-6, 2018. c⃝2018 Association for Computational Linguistics How Gender and Skin Tone Modifiers Affect Emoji Semantics in Twitter Francesco Barbieri LASTUS Lab, TALN Universitat Pompeu Fabra Barcelona, Spain name.surname@upf.edu Jose Camacho-Collados School of Computer Science and Informatics Cardiff University United Kingdom camachocolladosj@cardiff.ac.uk Abstract In this paper we analyze the use of emojis in social media with respect to gender and skin tone. By gathering a dataset of over twenty two million tweets from United States some findings are clearly highlighted after perform- ing a simple frequency-based analysis. More- over, we carry out a semantic analysis on the usage of emojis and their modifiers (e.g. gen- der and skin tone) by embedding all words, emojis and modifiers into the same vector space. Our analyses reveal that some stereo- types related to the skin color and gender seem to be reflected on the use of these modifiers. For example, emojis representing hand ges- tures are more widely utilized with lighter skin tones, and the usage across skin tones differs significantly. At the same time, the vector cor- responding to the male modifier tends to be semantically close to emojis related to busi- ness or technology, whereas their female coun- terparts appear closer to emojis about love or makeup. 1 Introduction Gender and race stereotypes are still present in many places of our lives. These stereotype-based biases are directly reflected on the data that can be gathered from different sources such as vi- sual or textual contents. In fact, it has been shown how these biases can lead to problematic behaviours such as an increase in discrimination (Podesta et al., 2014). These biases have already been studied in diverse text data sources (Zhao et al., 2017), and have been proved to propagate to supervised and unsupervised techniques learning from them, including word embeddings (Boluk- basi et al., 2016; Caliskan et al., 2017; Zhao et al., 2018) and end-user applications like online ads (Sweeney, 2013). In this paper we study the biases produced in a newer form of communication in social media from an analytical point of view. We focus on the use of emojis and their interaction with the textual content within a social network (i.e. Twitter). We study emojis as another part of the message, as it could be words. An interesting feature about emo- jis, apart from their increasing use in diverse social media platforms, is that they enable us to numeri- cally measure some biases with respect to gender and race. Recently, emojis have introduced mod- ifiers as part of their encoding. With these mod- ifiers the same emoji can be used with different features: as male or female, or with different skin colors, for example. We approach the problem from two method- ological perspectives. First, we analyze the use of emojis and their modifiers from a numerical point of view, counting their occurrences in a corpus. This already gives us important hints …

Content truncated to 500 words