A Review on Emotion Detection in Online Social Networks: Multi-Label Learning Approach

Emotion recognition in online social networks has grown sophisticated, particularly with the advent of multi-label emotions represented across varied content such as text, images, and videos . Such expressions usually manage to indicate more than one emotion at a time, with users in secluded online environments having greater likelihoods of misinterpretation. While AI models like BERT with multi-label classifiers have shown a maximum of 97.5% accuracy in emotion detection, they do so without label correlations and without live multi-modal interventions. Evidence also indicates that small interactive learning and context-aware features built into models are superior to static classification or single- labeling techniques in expressing emotional subtlety. In this work, we introduce a multi-stage multi-label learning paradigm to identify emotions within social networks by integrating AI-based identification with correlation-backed methods. Our model operates in four phases: Awareness Features, Micro-Emotion Indicators, Reinforcement Learning, and AI- Driven Multi-Label Classification. Collectively, these phases enable models to demonstrate multiple emotions, delay before labeling, and construct correct emotional profiles as time passes. This dual-modeling strategy seeks to enhance emotion detection as an engaging, personalized, and effective practice in real social network environments.