State-of-the-Art Machine Learning Techniques in Sentiment Analysis for Social Media

This research reviews state-of-the-art machine learning techniques for sentiment analysis in social media, tracing the evolution from traditional models like Support Vector Machines (SVM) and Naïve Bayes to advanced deep learning and transformer-based architectures. The evolution of the field is marked by the adoption of methods that leverage contextual embeddings, hybrid neural networks, and the integration of multimodal signals such as emojis, all of which have contributed to improved classification in diverse and noisy social media environments. Among recent advances, certain hybrid architectures optimized with algorithmic search have demonstrated superior performance, achieving accuracy and macro-F1 values approaching 97% on large-scale Twitter datasets. Some of them utilize fusion of textual and visual features with attention mechanisms, as well as models benefiting from transfer learning and automated feature selection, each excelling with macro-F1 scores in the range of 0.57–0.83 on challenging multilingual and specialized benchmark sets. This progress highlights the importance of multimodal integration, sophisticated preprocessing, and adaptive model design in addressing the variability and complexity inherent in global social media data.