Abstract

In this study, we aim to predict the most likely emoji given only a short text as an input. We extract a Hebrew political dataset of user comments for emoji prediction. Then, we investigate highly sparse n-grams representations as well as denser character n-grams representations for emoji classification. Since the comments in social media are usually short, we also investigate four dimension reduction methods, which associates similar words to similar vectorial representation. We demonstrate that the common Word Embedding dimension reduction method is not optimal. We also show that the character n-grams representations outperform all the other representation for the task of emoji prediction for Hebrew political domain.