Abstract
[This Proceedings paper was revised and published in the 2019 issue of the journal Issues in Informing Science and Information Technology, Volume 16] Aim/Purpose: Any system that aims to address the task of modeling social media communication need to deal with the usage of emojis. Efficient prediction of the most likely emoji given the text of a message may help to improve different NLP tasks. Background: We explore two tasks: emoji identification and emoji prediction. While emoji prediction is a classification task of predicting the emojis that appear in a given text message, emoji identification is the complementary preceding task of determining if a given text message includes emojies. Methodology: We adopt a supervised Machine Learning (ML) approach. We compare two text representation approaches, i.e., n-grams and character n-grams and analyze the contribution of additional metadata features to the classification. Contribution: The task of emoji identification is novel. We extend the definition of the emoji prediction task by allowing to use not only the textual content but also meta-data analysis. Findings: Metadata improve the classification accuracy in the task of emoji identification. In the task of emoji prediction it is better to apply feature selection. Recommendations for Practitioners: In many of the cases the classifier decision seems fitter to the comment con-tent than the emoji that was chosen by the commentator. The classifier may be useful for emoji suggestion. Recommendations for Researchers: Explore character-based representations rather than word-based representations in the case of morphologically rich languages. Impact on Society: Improve the modeling of social media communication. Future Research: We plan to address the multi-label setting of the emoji prediction task and to investigate the deep learning approach for both of our classification tasks.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have