Federated Learning-Based Architecture for Personalized Next Emoji Prediction for Social Media Comments

Abstract

In the era of digital communication, emojis have become the vibrant palette of people’s textual expressions, adding depth and emotion to messages. However, deciphering the subtle nuances of emojis poses a unique challenge due to their inherent ambiguity. The quest for predicting the next emoji in technological devices has emerged at the forefront of predictive analytics, demanding the analysis of extensive and diverse datasets while respecting user privacy. Enter Federated Learning (FL), a groundbreaking approach that thrives on learning from decentralized data sources without compromising confidentiality. This study delves into the unexplored realm of Federated Learning-based emoji prediction. Utilizing a tailored adaptation of BERT with a rich corpus of text (drawn from social media), encompassing both words and emojis, the author’s innovative architecture aims to predict the most fitting emoji for a given text while prioritizing user privacy. Welcome to EmojiSculpt, where the future of personalized emoji prediction takes center stage.

Keywords

• Emoji prediction
• Bengali language
• privacy-preserving
• federated learning
• BERT
• sentiment