Hybrid IndoBERT–LLaMA for Aspect-based Sentiment Analysis of Indonesian Mobile Banking Reviews
User reviews of mobile banking applications provide valuable insights for service improvement, yet the absence of labeled Indonesian datasets limits effective sentiment analysis. The study applies Aspect-Based Sentiment Analysis (ABSA) using a hybrid of IndoBERT and LLaMA, combining IndoBERT’s contextual sensitivity with LLaMA’s reasoning ability. A dataset of 5,513 reviews (January–June 2025) was fine-tuned and evaluated. The hybrid ensemble achieved 0.950 accuracy and 0.862 F1-Macro, outperforming single models by +0.017 and +0.041 respectively. The gains demonstrate the effectiveness of hybrid deep learning for Indonesian sentiment analysis. Beyond quantitative improvements, the framework also supports adaptive user experience through deep learning and enhances emotional intelligence in human–AI collaboration, enabling more empathetic and user-centric digital financial services.