<![CDATA[This study investigates user sentiment toward leading Indonesian e‑commerce applications through deep‑learning‑based text classification. A balanced corpus of 50,000 Indonesian‑language reviews was collected from Google Play and App Store for Tokopedia, Shopee, Bukalapak, Lazada, and Blibli. We applied two state‑of‑the‑art approaches—Long Short‑Term Memory (LSTM) networks enriched with pre‑trained FastText embeddings and fine‑tuned Bidirectional Encoder Representations from Transformers (BERT; IndoBERT v2). Data pre‑processing included text cleaning, slang normalization, stemming, and tokenization following the KBBI standard. Both models were trained with an 80:20 stratified split and evaluated using accuracy, precision, recall, F1‑score, and AUC. BERT achieved 90.6 % accuracy and 90.1 % F1‑score, outperforming LSTM's 83.2 % accuracy and 82.7 % F1‑score. McNemar’s test indicated the improvement is statistically significant (p < 0.01). These findings show that contextual embeddings capture nuanced Indonesian sentiments more effectively than sequential RNN‑based approaches, offering actionable insights for e‑commerce stakeholders to enhance customer experience.]]>
Copyrights © 2025
