Classifying the Informative Behaviour of Emoji in Microblogs

Classifying the Informative Behaviour of Emoji in Microblogs Giulia Donato∗, Patrizia Paggio∗† ∗University of Copenhagen giulia.dnt@gmail.com, paggio@hum.ku.dk †University of Malta patrizia.paggio@um.edu.mt Abstract Emoji are pictographs commonly used in microblogs as emotion markers, but they can also represent a much wider range of concepts. Additionally, they may occur in different positions within a message (e.g. a tweet), appear in sequences or act as word substitute. Emoji must be considered necessary elements in the analysis and processing of user generated content, since they can either provide fundamental syntactic information, emphasize what is already expressed in the text, or carry meaning that cannot be inferred from the words alone. We collected and annotated a corpus of 2475 tweets pairs with the aim of analyzing and then classifying emoji use with respect to redundancy. The best classification model achieved an F-score of 0.7. In this paper we shortly present the corpus, and we describe the classification experiments, explain the predictive features adopted, discuss the problematic aspects of our approach and suggest future improvements. Keywords: Emoji, Microblogs, Redundancy, Supervised Learning 1. Introduction Emoji are non verbal features used to enrich computer me- diated communication (CMC) and mobile mediated com- munication (MMC). Empirically, the use of emoji may vary in non trivial ways: emoji can be redundant with respect to the text, but they can also act as words, carrying their own part-of-speech category, thus providing fundamental semantic information. Studies on emotion expression in text have noted that emotional emoji may carry information which could not be inferred from the words alone. Follow- ing these findings, this work proposes to further investigate the informative behaviour of emoji in microblogs. Recognizing to what extent emoji are redundant with re- spect to one or more words in the text could be helpful for further research in automatic content labeling; under- standing how (and to what extent) emoji convey additional meaning or have a syntactic function can be important to improve the results in other NLP tasks such as metaphor detection and text summarization. 2. Literature Emoji are picture characters, or pictographs, initially de- veloped by Shigetaka Kurita during the late nineties. The initial set was further expanded and eventually became part of the Unicode standard in 2009 (Kelly, 2015; Miller et al., 2016). Emoji in CMC are, similarly to emoticons, mostly used to express emotions; according to (Swiftkey, 2015) the top used emoji categories are the ones that include the happy and the sad faces. Novak et al. (2015) confirm that these preferences apply also to Twitter users. While Boia et al. (2013) demonstrate that emoticons are not necessarily ac- curate in retrieving sentiment words from a corpus, Halls- mar and Palm (2016) show that an emoji heuristic can be ef- fectively used to retrieve tweets corresponding to a specific emotion class (within a multiclass framework), perhaps in- dicating that emotional emoji to some extent may behave redundantly with respect to the tweet text. Since both emoticons and emoji can be employed with a wide range of purposes …

Content truncated to 500 words