Incorporating Emoji Descriptions Improves Tweet Classification

Proceedings of NAACL-HLT 2019, pages 2096–2101 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 2096 Incorporating Emoji Descriptions Improves Tweet Classification Abhishek Singh, Eduardo Blanco and Wei Jin Computer Science and Engineering University of North Texas Denton, TX, 76203 abhisheksingh@my.unt.edu, {eduardo.blanco,wei.jin}@unt.edu Abstract Tweets are short messages that often include specialized language such as hashtags and emojis. In this paper, we present a simple strategy to process emojis: replace them with their natural language description and use pretrained word embeddings as normally done with standard words. We show that this strategy is more effective than using pretrained emoji embeddings for tweet classification. Specifically, we obtain new state-of-the-art results in irony detection and sentiment analysis despite our neural network is simpler than previous proposals. 1 Introduction Tweets are short messages shared on Twitter, one of the most popular social networking ser- vices with 326 million monthly active users word wide (Twitter, 2018). Tweets often use special- ized language such as abbreviations (e.g., TBH: To be honest), hashtags (e.g., #NBAFinals), emoti- cons and emojis. The Oxford Dictionary defines an emoticon as “a facial expression such as a smile or frown, formed by various combinations of key- board characters” (e.g., “:)”, “:-(”), and an emoji as “a small digital image or icon used to express an idea or emotion” (e.g., , , ). While the number of emoticons is relatively small, the Uni- code Standard includes over 2,800 emojis. Emojis are interesting because they succinctly encode meaning that otherwise would require more than one word to convey (e.g., grinning face, clapping hands and face with medical mask for the emojis above). Additionally, emojis have become popular in social media. 5 billion emojis are sent daily on Facebook (Burge, 2018). While only 6% of the top-100 Facebook headlines used emojis in 2015, 52% did so in 2017 (Boland, 2017). Over Irony? I just love being ignored |#not Yes Love it when my mans on a clean- ing spree.. Saves me doing it No Sentiment @Paul OConnor187 hi we going to see ted 2 at the Odeon cinemas at Glasgow on Wednesday Positive Serato DJ isn’t compatible with Windows 10 yet ...got to spin on my old laptop Saturday. Negative Table 1: Sample tweets with their irony and sentiment judgements. Note that the emojis help to determine irony usage and the author’s sentiment. 14% of tweets and 50% of Instagram posts con- tain at least one emoji (Cruse, 2015; Moon, 2015). Irony detection and sentiment analysis in tweets are two popular tasks. Sentiment analysis has re- ceived substantially more attention than irony de- tection. Irony, however, is a major error source in sentiment analysis (0.71 F1 overall but 0.29 F1 with ironic tweets (Hee et al., 2018)), and natural language understanding in general does not gener- alize well with ironic texts (Liu et al., 2012; May- nard and Greenwood, 2014). In this paper, we tackle both irony and sentiment analysis in tweets—two classification tasks. In particular, we focus on modeling emojis. …

Content truncated to 500 words