Advances in Emotion Recognition through Twitter Text: A Review of Ensemble and Novel Techniques

The rise of social media platforms, especially Twitter, has positioned textual emotion recognition as a key component in understanding public opinion, behavioural dynamics, and emerging social trends. Given Twitter's rapid pace and informal nature characterized by abbreviations, slang, code-switching, and sarcasm ₁₁accurate emotion detection from tweets presents a complex challenge. Yet, the platform's immense volume of user-generated content provides an unparalleled opportunity for real-time emotional analysis at scale. This review critically surveys recent advancements in tweet-based emotion recognition, with a particular emphasis on novel machine learning techniques and ensemble-based frameworks. It explores how various computational models including deep learning and transformer architectures are applied to extract emotional signals from short texts. In addition, we compare widely used benchmark datasets, discuss text representation techniques, and evaluate the trade-offs between standalone and ensemble classifiers in capturing emotional depth and diversity. Beyond model performance, the paper also addresses the growing importance of context-aware learning, multimodal integration (e.g., emojis, hashtags, and user metadata), and transfer learning across linguistic and cultural boundaries. Persistent challenges such as limited labelled data, lack of model transparency, and the need for real-time deployment are highlighted. The paper concludes by outlining future research directions aimed at enhancing the adaptability, interpretability, and scalability of emotion recognition systems. This review serves as a foundational guide for researchers, developers, and practitioners in the domains of affective computing and social media analytics.