SVSD: A Comprehensive Framework for Vietnamese Sentiment Analysis

Sentiment analysis is essential for understanding public opinion, customer feedback, and social media interactions, providing valuable insights for businesses and researchers. Despite recent advancements in Vietnamese sentiment analysis, high accuracy often comes at the cost of computationally intensive models, limiting scalability and real-time application in resource-constrained environments. Challenges like data imbalance and extensive preprocessing further underscore the need for more efficient methods. This study evaluates sentiment classification in Vietnamese using the PhoBERT tokenizer and FastText representation with machine learning models across three datasets: UIT-VSFC, Web Content, and UIT-VSMEC. We introduce SVSD (SVM-based Vietnamese Sentiment Detection), which combines PhoBERT preprocessing, FastText embeddings, and SVM classification. SVSD achieved top accuracy scores of 94% on UIT-VSFC, 95% on Web Content, and 93% on UIT-VSMEC, demonstrating its effectiveness and highlighting the benefits of specialized preprocessing tools like PhoBERT for Vietnamese sentiment analysis.