Emoji Prediction from Arabic Sentence
Takua Mokhamed, Ashraf Elnagar · 2022
Emoji prediction aims to identify the appropriate emoji or set of emojis associated with the text. With the help of a machine and deep learning, models can learn and comprehend rich representations of the contextual and emotional intent of written texts. However, each emoji has a large conceptual and emotional range. Accordingly, each emoji may be used in a variety of sentences. We can choose the proper emoji for each text by assessing its content and attitude. As a result, one of the challenges that have drawn academics' attention is predicting an emoji from a given textual input. Even though the majority of existing research on the emoji prediction task focuses on the English language, no study has been done on the Arabic language in the emoji prediction task from a whole sentence. Therefore, in this project, using deep learning LSTM, an attempt for the first time has been made to predict the emoji for Arabic text. Similarly, three machine learning algorithms were examined, and the performance of the models was analyzed. Moreover, we were able to achieve, with a deep learning model based on LSTM, an accuracy of 80% using the 20 most common emojis, which is much greater than the accuracy of the SVM, Naive Bayes, and Random Forest models, which were respectively 29%, 27%, and 26 %. Furthermore, due to the unavailability of an Arabic emoji prediction dataset, the dataset has been reconstructed by translating the text in the existing English emoji-text pair dataset into Arabic text.