A Multi-Model Approach Combining BERT and Generative AI

Social media (Twitter) is an interactive platform where people and businesses voice opinions on matters, hence making it a good source of gathering insights about human behavior, market trends, and social dynamics. Sentiment classification of tweets as positive, negative, or neutral using categorization via machine learning and sophisticated natural language processing (NLP) methods is the subject of this research. A comparative study was performed between eight machine learning models: Logistic Regression, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree Classifier, Random Forest Classifier, Bagging Classifier, Extra Trees Classifier, and AdaBoost Classifier. Of these, SVM had the best performance with a test accuracy of 97.46%, as it can handle high-dimensional text features and can clearly define decision boundaries. Ensemble techniques such as Random Forest and Extra Trees were also quite strong but not as strong as expected, which could be due to overfitting or the poor choice of hyperparameters. For this paper, we applied a fine-tuned BERT-based Generative AI model with 98.19% accuracy for sentiment classification, greatly increasing the accuracy in classification. By means of this transformer-based framework, the model could create a more contextualized understanding of language and manage linguistic complexity, sarcasm, and ambivalence much more effectively than the conventional machine learning methods. Such findings reflect the shift from the standard machine learning models to Generative AI when it comes to sentiment analysis, as well as its possible superiority over conventional methods in real-world applications.