Fine Grained Emotion Detection On Twitter Using Transformer -Based Deep Learning Models

Abstract

Recent years have seen a proliferation of users expressing their feelings on social media sites like Twitter, which has opened the door for real-time sentiment analysis.  By distinguishing between distinct emotional states including happiness, fury, sorrow, fear, and surprise, fine-grained emotion recognition surpasses the capabilities of simple sentiment analysis.  This research introduces a deep learning method that uses transformers to identify emotions on Twitter with a finer degree of specificity.  The system is able to extract intricate linguistic patterns and contextual relationships from brief and noisy social media material by using state-of-the-art models like BERT and RoBERTa.  The model is trained and tested using a dataset of pre-processed tweets that have been tagged with emotions.  The experimental findings show that when compared to typical machine learning baselines, there is a considerable gain in classification accuracy and F1-score.  The suggested model is suitable for practical emotion mining tasks because to its resilience in dealing with language ambiguity, sarcasm, and code-mixed information.  Possible uses for this study include social behavior modeling, customer feedback analysis, and mental health monitoring. It also helps to increase AI systems' emotional intelligence.