earticle

영어

The exponential growth of online information has posed significant challenges for analyzing and classifying user sentiments, especially in complex linguistic contexts. This paper introduces a novel approach combining Ontology and a fine-tuned Transformer model (BERT) for automated sentiment analysis. Unlike traditional methods, our approach integrates a domain-specific Ontology to enhance contextual understanding and semantic reasoning, addressing challenges like sarcasm and ambiguous expressions. Experimental results on diverse datasets, including Vietnamese social media comments and international sentiment datasets, demonstrate the superiority of our system, achieving a 95.64% Quality of Experience (QoE) score―significantly outperforming existing methods. This study not only advances sentiment analysis techniques but also provides insights into improving user satisfaction across various applications, including customer feedback analysis, education, and public opinion monitoring. Our results indicate higher label-wise precision (QoE) on Vietnamese social comments; we also provide brief qualitative examples illustrating when the ontology helps.