Sentiment Synthesis Transforming YouTube Comments into Strategic Insights

Abstract

Sentiment analysis in the domain of social media analytics is crucial for understanding public opinion, particularly on platforms like YouTube, where vast amounts of user-generated content are continuously uploaded. The challenge lies in efficiently processing live-streamed comments in real time while ensuring accurate classification into sentiment categories—positive, negative, and neutral. Existing approaches, such as those utilizing BERT-based models, face limitations in handling multilingual and code-mixed comments effectively, leading to reduced accuracy in sentiment classification. The proposed system enhances sentiment analysis system leveraging XML-RoBERTa, a transformer-based model trained on diverse multilingual datasets, improving classification accuracy and robustness. The system processes live YouTube comments in real time, categorizes them into sentiment classes, and visualizes the results using word clouds, pie charts, and bar charts. Additionally, it generates a detailed sentiment report that maps individual comments to their respective sentiment categories, offering a transparent and comprehensive analysis. Experimental results demonstrate that XML-RoBERTa outperforms the BERT model used in the base paper by offering better multilingual support and sentiment differentiation, validating the effectiveness of the proposed approach.