So, me and my friend was participating IPavlov course on deep learning in NLP. As out final project we want to work on some sort of new, unresolved problem That's why we decided to train model for predicting how to put emojis around some text. Occasionally, the biggest publicly available source of emojized text is Twitter. We open this work to public, so everyone can work on model design, not data preparation.

Content

The original files for this dataset was four archives from ArchiveTeam TwitterStream project. We reformatted these files, selected all the english-language tweets with at least one emoji. Also there was some ordinary preprocessing: removing hashtags, urls, mentions.