Sentiment Analysis Using Transformer Based Hybrid Models

Sentiment Analysis represents a fundamental component of Natural Language Processing (NLP), which detects emotional content in digital text, offering positive, negative, and neutral categories. The system enables businesses to analyze customer opinions and detect brand reputation through market trend assessment, which supports improved decision-making processes. Such applications utilize sentiment analysis across ecommerce platforms, customer services, social media websites, and financial sectors to gain important customer feedback. Our paper presents a decision-level, precedence-rule hybrid that combines T5 with RoBERTa. Sentiment analysis from textual data using RoBERTa relies on transformer technologies that were specifically optimized for sequence classification. Through its text-to-text operation, T5 produces human-like sentiment classifications that appear as prompts within its responses. The models enable precise analysis of large datasets through their improved accuracy, scalability, and contextual understanding features. On the Twitter data, the hybrid consistently outperforms standalone models, improving neutrality resolution and achieving accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 6. 0 1 \%}$</tex>, precision of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 6. 1 3 \%}$</tex>, recall of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 6. 0 0 \%}$</tex> and F1-score of 96.05%.