Informasi Lainnya
Abstraksi
User reviews of energy service applications contain multiple aspects and sentiments with imbalanced distributions, which complicates automated analysis. This study applies Aspect-Based Sentiment Analysis (ABSA) using the IndoBERTweet model to analyze user reviews of PT XYZ’s application for aspect and sentiment classification tasks. Several imbalance-handling strategies are evaluated, including a baseline model, SMOTE, Easy Data Augmentation (EDA), and SMOTE–Tomek. The results show that IndoBERTweet effectively captures the context of Indonesian user reviews; however, its performance is influenced by data imbalance, particularly for minority aspects and neutral sentiment. For both aspect and sentiment classification, EDA achieves the most stable performance and best generalization on the test data compared to resampling-based methods. Error analysis indicates that neutral sentiment and overlapping aspects remain the main challenges due to linguistic ambiguity. This study concludes that text-based augmentation is more effective for handling imbalanced data in ABSA tasks on real-world user reviews.
- CII7G3 - PEMROSESAN BAHASA ALAMI LANJUT
- CAK79FB4 - PENGOLAHAN BAHASA ALAMI LANJUT
Koleksi & Sirkulasi
Tersedia 1 dari total 1 Koleksi
