Generalizing sentiment analysis: a review of progress, challenges, and emerging directions

Abstract Sentiment analysis is an increasingly vital technique within natural language processing for interpreting human emotions expressed in text. This survey explores the trajectory of sentiment analysis research, examining advancements from traditional machine learning approaches to state-of-the-art deep learning models, including Transformers and hybrid architectures. We highlight key challenges such as domain adaptation, linguistic diversity, and the evolving nuances of digital communication. This review distinguishes itself by adopting a multidisciplinary approach, integrating advancements from machine learning, cognitive science, and linguistics to address generalization, multimodal data integration, and the potential of self-supervised learning. Unlike prior surveys, our work provides a comprehensive synthesis of recent and emerging methodologies, although introduced in previous literature, remain scattered across domain specific studies such as hybrid models combining RoBERTa-GRU and Capsule Networks with semantic rules, while emphasizing ethical considerations and novel directions like adaptive feature selection and fairness-aware training. By providing comprehensive insights into applications across domains like healthcare, finance, and disaster management, this survey serves as a foundational resource for the next generation of sentiment analysis tools.