Persian Emoji Prediction Using Deep Learning and Emoji Embedding

Abstract

The appearance of social networks and the increasing expansion of these networks has created many challenges, especially in the field of natural language processing. One of these social networks that has been welcomed by many researchers is Twitter. Twitter’s users have the opportunity to consider one or more emojis for a tweet depending on the feeling and meaning of the tweet. Emojis contain information and concepts that the author of each tweet has in mind, the semantic and emotional range of each emoji is very wide and each emoji can be used in many different types of sentences. Therefore, by analyzing the content and emotion of each tweet, we can achieve the appropriate emoji of that tweet. For such reasons, predicting an emoji for a textual data is one of the challenges that has attracted the attention of researchers. In this article, using deep neural networks an attempt for the first time has been made to predict the emoji for Persian text data extracted from Twitter. And we were able to achieve F-score of 33% in 10 most frequent emojis which is 5% higher than the result of the SVM model and also 11% better than the result of the Naïve Bayes model, and F- score of 46% in 5 most frequent emojis which is 5% higher than the result of the SVM model and also 5% better than the result of the Naïve Bayes model.