RIP Emojis and Words to Contextualize Mourning on Twitter
Date
2021
Authors
Xu, Xinyuan
Manrique, Ruben
Pereira Nunes, Bernardo
Journal Title
Journal ISSN
Volume Title
Publisher
Association for Computing Machinery (ACM)
Abstract
This paper aims to investigate the use of emojis to contextualize mourning on Twitter. Specifically, we seek to determine (i) whether an emoji is sufficient to contextualize expressions of grief; (ii) which emojis most accurately represent mourning; (iii) whether only words are used to contextualize mourning; (iv) which words are used to characterize mourning in tweets; and, (v) if there are differences in the expression of mourning in different languages. For this, we use a multi-stage method to conduct a comprehensive analysis of the manifestations of grieving behavior on Twitter, and created machine learning models to classify expressions of mourning in tweets. The main contributions from this work are (1) a gold standard of manually annotated mourning tweets; (2) classification models produced using machine learning ensemble methods and BERT contextual embeddings; and, (3) an extensive analysis of our findings opening up opportunities for new research. The results of this paper reveal emojis alone are insufficient for identifying expressions of mourning in tweets, and the combination of both emojis and words is the most effective strategy for contextualizing mourning online-the models achieved the 84.8%-97% F1 score in all datasets. Although words alone are capable of characterizing mourning contexts correctly, the English vocabulary is limited, and the contribution of RIP-the abbreviation for "rest in peace''-is highly decisive. Our results have also shown that the most relevant emojis for this context were emotional ones, such as \includegraphics[width=1em]twitter_brokenheart.png, and emojis are used in a uniform fashion in both Spanish and English.
Description
Keywords
emojis, social media mourning, classification models, BERT
Citation
Collections
Source
Type
Conference paper
Book Title
Entity type
Access Statement
License Rights
Restricted until
2099-12-31