Rancang Bangun Sistem Analisis Sentimen Opini Publik Multi-Source Berbasis Pemahaman Konteks Menggunakan Fine-Tuning IndoBERT (Studi Kasus Universitas Negeri Padang)

This study aims to develop an interactive, web-based, multi-source sentiment analysis system for public opinion to facilitate the monitoring of societal perception regarding Universitas Negeri Padang. The methodology integrates a Transformer-based natural language processing model—specifically, a fine-tuned IndoBERT—executed through a client-server architecture (Next.js and FastAPI) and enhanced with a hybrid inference logic mechanism (lexicon-based override) for batch data processing. Evaluation results utilizing a confusion matrix indicate that the IndoBERT model achieved an accuracy of 95% in classifying 3,411 rows of test data, with a peak precision of 97% for positive sentiment and a recall of 96% for negative sentiment. The infrastructure implementation proved reliable in handling text-computational loads from diverse digital platforms (X, Instagram, and news portals) without latency issues. Consequently, this innovation establishes a scalable, accurate, and adaptive Decision Support System (DSS) solution, providing an objective foundation for institutional evaluation and decision-making.