Multilingual Transformer Contextual Embedding Model for Political Tweets Analysis

In India, the rise of social media usage has led to significant political discourse on platforms such as Twitter (now known as X), creating new opportunities for political sentiment analysis. With these new opportunities, still the challenges such as multilingualism, code-switching, sarcasm, and regional dialects limit the effectiveness of sentiment analysis methods. In order to address these challenges, this paper introduces a novel Multilingual Transformer Contextual Embedding Model to handle the complexities of Indian political tweets. We train and evaluate the model on a dataset of 3 million political tweets in Hindi, English, and regional languages. The results demonstrate an accuracy of 98.5%, surpassing existing state-of-the-art models by an average of 2.5%. Additionally, the introduced model achieved an F1 score of 97.5%, precision of 98%, and recall of 97%, showcasing its superiority in handling multilingual and context-specific nuances. Comparative analysis indicates that transformer-based models like the proposed model significantly outperform traditional machine learning models in sentiment classification and bias detection. The proposed model can be enhanced with the real-time processing capabilities, more multilingual support, and ethical considerations for more inclusive and unbiased political sentiment analysis frameworks.