Implementation of Recurrent Network for Emotion Recognition of Twitter Data

Abstract

A new generation of emoticons, called emojis, is being largely used for both mobile and social media communications. Emojis are considered a graphic expression of emotions, and users have been widely used to express their emotions in social media. Emojis are graphic unicode symbols used to express perceptions, views, and ideas as a shorthand. Unlike the small number of well-known emoticons carrying clear emotional content, hundreds of emojis are being used in different social networks. The task of emoji emotion recognition is to predict the original emoji in a tweet. Recurrent neural network is used for building emoji emotion recognition system. Glove is a word-embedding method used for obtaining vector representation of words and are used for training the recurrent neural network. This is achieved by mapping words into a meaningful space where the distance between words is related to semantic similarity. Based on the word embedding in the Twitter dataset, recurrent neural network builds the model and finally predicts the emoji associated with the tweets with an accuracy of 83%.