A Unified Hybrid Framework for Fine-Grained Emotion classification: Machine Learning–Deep Learning Synergy for Robust Performance

Abstract

Emotion Classification from tweets is a very important Natural Language Processing task that aims at automatically detecting the emotions from social media platforms. Social media posts are the real-time reflection of public sentiments and emotional well-being. Oftentimes, people express their emotions through unstructured posts such as tweets and comments. This research work makes a comparative analysis of the effectiveness of Machine learning and Deep Learning approaches for fine-grained Emotion classification. Traditional ML approaches like Logistic Regression, SVM, Naive Bayes are evaluated using TF-IDF and lexicon-based features. To capture the contextual semantics, deep learning models such as LSTM, CNN and transformer based models were evaluated. We also experimented with extensive hyperparameter optimization using grid search and bayesian tuning methods. Furthermore a novel hybrid ML-DL architecture is proposed to leverage the strength of both the paradigms. The experimental results demonstrate that the hybrid framework outperformed standalone methods in terms of accuracy, robustness and generalizability. The findings of the research work with 95.4% accuracy concludes that combining the shallow and deep feature spaces with optimized hyperparameters is essential for building a more reliable and scalable emotion classification system.