The increasing volume of consumer reviews on e-commerce platforms has highlighted the need for sentiment analysis methods capable of capturing user opinions more specifically concerning particular aspects of products or services. Aspect-Based Sentiment Analysis (ABSA) addresses this need by identifying the aspects discussed in a review and determining the polarity of sentiment expressed toward each aspect. This study aims to explore and compare the effectiveness of two deep learning models, namely Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) in performing ABSA on Indonesian-language e-commerce user reviews. The research methodology comprises several stages: data exploration and cleaning, text preprocessing, aspect and sentiment annotation, training of CNN and LSTM models, and performance evaluation using metrics such as accuracy, precision, recall, and F1-score. The dataset is divided into training, validation, and testing subsets. The analyzed aspects include delivery, product, price, application, and service. Results show that the LSTM model outperforms CNN across all evaluation metrics. LSTM achieved an accuracy of 86.10%, precision of 85.70%, recall of 85.90%, and an F1-score of 85.80%, while CNN reported slightly lower values. Based on these findings, LSTM proves to be more effective in understanding the contextual and linguistic structure of the Indonesian language in ABSA tasks. This study provides a valuable contribution to the development of automatic sentiment analysis systems in the e-commerce sector. Future research can expand this approach by incorporating transformer-based models such as IndoBERT or integrating attention mechanisms to further improve predictive accuracy. These findings offer practical insights for industry stakeholders seeking to enhance customer experience through a deeper understanding of user sentiment.
Adhelina et al. (Thu,) studied this question.