Understanding public perceptions of waste management: A comparative sentiment analysis using LSTM and IndoBERT

Abstract

Background: Waste management remains one of Indonesia’s pressing environmental and social challenges, characterized by increasing waste generation and public dissatisfaction with existing waste-treatment systems. Understanding public sentiment toward waste-management policies and community-based programs is essential for supporting more effective interventions. This study aims to examine public perceptions of waste-management issues in Indonesia and compare the performance of Long Short-Term Memory (LSTM) and Indonesian Bidirectional Encoder Representations from Transformers (IndoBERT) for sentiment classification. Methods: This study used a dataset of 15,000 public text opinions collected from social media using a crawling technique. The data were manually annotated into three sentiment categories: positive, neutral, and negative. After preprocessing, two sentiment-classification models, LSTM and IndoBERT, were developed and evaluated. Model performance was assessed using accuracy, precision, recall, and F1-score. Topic-based sentiment analysis was also conducted to identify public perceptions of major waste-management issues. Findings: The results showed that IndoBERT outperformed LSTM in sentiment classification by capturing contextual meaning and sentiment nuances in Indonesian-language expressions more effectively. Topic-based analysis indicated that community-based initiatives, including waste banks and recycling programs, were generally associated with positive sentiment. In contrast, landfill-related issues were more frequently associated with negative sentiment due to concerns about pollution and poor management. Conclusion: IndoBERT provides a more reliable approach for analyzing public sentiment toward waste-management issues in Indonesia. The findings highlight the importance of understanding public perceptions to strengthen community engagement, improve public communication, and support more effective waste-management policies and programs. Novelty/Originality of this article: The novelty of this study lies in integrating deep learning-based sentiment classification with topic-based sentiment analysis to examine public opinion specifically regarding waste-management strategies in Indonesia. The comparative evaluation of LSTM and IndoBERT also provides empirical insight into the suitability of contextual language models for analyzing Indonesian-language public opinions on waste management.