EMOJI PREDICTION IN TEXT-BASED COMMUNICATION: A STUDY OF MACHINE LEARNING APPROACHES

Abstract

Because we use digital technologies for communication, emojis are important for making our emotions, intentions and other details clear. This paper looks at emoji prediction as a way to classify text and recommend or add suitable emojis in it using Natural Language Processing. We review and test different machine learning algorithms, networks and language models such as Naive Bayes, support vector machines, LSTMs, BiLSTMs, Transformer-based models, BERT and RoBERTa. By using the datasets from Twitter and Reddit, we study the accuracies, precisions, recalls and F1-scores of these models. We found that transformer-based models in the context-aware category are most effective in detecting the semantic and emotional cues for predicting the right emoji. In addition, we review the effects of using emojis in different cultures and on various platforms, consider challenges with understanding sarcasm and ambiguity and introduce the idea of moving toward multimodal systems. Through its findings, this study enhances how affective computing is used and how users feel when interacting with chatbots, automated messaging and social media.

Keywords: Emoji Prediction can be done using Natural Language Processing (NLP), Text Classification, Machine Learning, Media Analysis, Multimodal Learning, Context-Aware Models and Emotion Recognition