Hatching Chick at SemEval-2018 Task 2: Multilingual Emoji Prediction

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 445–448 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics Hatching Chick at SemEval-2018 Task 2: Multilingual Emoji Prediction J. Coster, R.G. van Dalen, and N.A.J. Stierman Department of Information Science University of Groningen {j.coster.2, r.g.van.dalen, n.a.j.stierman}@student.rug.nl Abstract As part of a SemEval 2018 shared task an at- tempt was made to build a system capable of predicting the occurence of a language’s most frequently used emoji in Tweets. Specifically, models for English and Spanish data were created and trained on 500.000 and 100.000 tweets respectively. In order to create these models, first a logistic regressor, a sequential LSTM, a random forest regressor and a SVM were tested. The latter was found to perform best and therefore optimized individually for both languages. During developmet f1-scores of 61 and 82 were obtained for English and Spanish data respectively, in comparison, f1- scores on the official evaluation data were 21 and 18. The significant decrease in perfor- mance during evaluation might be explained by overfitting during development and might therefore have partially be prevented by using cross-validation. Over all, emoji which occur in a very specific context such as a Christmas tree were found to be most predictable. 1 Introduction It is said that a picture is worth a thousand words; inherently then, visual icons can provide addi- tional meaning to text. One common example of this is the use of emoticons (emoji) accompanying primarily short, informal texts such as text mes- sages and tweets. The role of these is interesting as they can be used in a variety of manners such as to complement text (e.g., to indicate happiness) and to replace text (e.g., I you instead of I love you). This research is concerned with the develop- ment of a system for predicting the occurrence of these emoji on the basis of Twitter data and is conducted as part of a SemEval 2018 shared task (Barbieri et al., 2018). More specifically, given the text of an English or Spanish tweet, the system will attempt to predict the emoji originally found in that tweet. The set of emoji used is limited to the twenty most frequent emoji for each language respectively. 2 Previous Work A similar research on predicting emoji occurring in tweets was conducted by Barbieri et al. (2017). In this study a Bidirectional Long Short-Term Memory (BLSTM) network with standard look- up word representations and character-based rep- resentations of tokens was used. While it was found that profoundly dissimilar emoji could be predicted, this method did not succeed to accu- rately differentiate between twenty emoji classes. Na’aman et al. (2017) meanwhile conducted a study to further explore the linguistic varieties of the purposes of emoji on Twitter. It was found that emoji can be an integral part of the content. One common example of this is a part of a phrase be- ing replaced by an emoji, much like the ’I you’ example from the previous …

Content truncated to 500 words