Bidirectional GRU for Aspect-Based Sentiment Classification in Multi-Dimensional Review Analysis

Sri Redjeki (Universitas Teknologi Digital Indonesia)
Basanto Joshi (Tribhuvan University)
Alfonso Situmorang (Universitas Methodist Indonesia)
Muhammad Guntara (Universitas Teknologi Digital Indonesia)
Sri Rezeki Candra Nursari (Universitas Pancasila)
Dara Kusumawati (Universitas Teknologi Digital Indonesia)

Zero : Jurnal Sains, Matematika, dan Terapan
Vol 9, No 2 (2025): Zero: Jurnal Sains Matematika dan Terapan

Traditional markets in Yogyakarta face mounting pressure from modernization and digital retail competition, yet user-generated reviews remain underutilized. This study applies Aspect-Based Sentiment Analysis (ABSA) with a Bidirectional Gated Recurrent Unit (BiGRU) on 9,222 annotated reviews from nine markets (2016-2024). BiGRU was chosen not only for its efficiency but also for its robustness in low-resource, multilingual settings with informal expressions, where transformer models often require larger datasets and compute. The best configuration with 64 GRU units and a 70:15:15 split achieved 83.4% accuracy (95% CI: ±1.2%) and an F1-score of 0.813, surpassing baselines such as Na