SPAN-LEVEL ASPECT-BASED SENTIMENT TRIPLET ANALYSIS IN GOVERNMENT APPLICATION REVIEWS

Feza Raffa Arnanda (Politeknik Statistika STIS)
Lya Hulliyyatus Suadaa (Politeknik Statistika STIS)
Avi Rudianita Indah Dg Widya (Politeknik Statistika STIS)
Setia Irham Pramana (Politeknik Statistika STIS)

AL-ULUM: JURNAL SAINS DAN TEKNOLOGI
Vol 12, No 1 (2026)

The government is enhancing digital public services through mobile applications in line with the Electronic-Based Government System (SPBE) 2018–2025 vision. To support continuous innovation, the Ministry of Administrative and Bureaucratic Reform (Kemenpan-RB) organize the Public Service Innovation Competition (KIPP). Understanding user complaints is essential, and aspect-based sentiment analysis, particularly Span-Level Aspect Sentiment Triplet Extraction (Span-ASTE), was applied to analyze government app reviews. A domain-specific dataset was developed with a Cohen’s Kappa of 0.817, indicating strong annotation reliability. IndoBERT-large achieved the highest F1-score of 0.76, while IndoBERT-lite-base provided an efficient alternative with an F1-score of 0.727. An aspect categorization model reached 0.86 accuracy. These models aim to improve public services, strengthen SPBE implementation, and enhance Indonesia’s E-Government Development Index ranking.