Early Detection of Environmental Issues from Social Media using IndoBERT and LDA: Case Study of Pollution and Deforestation in Indonesia

This study proposes a method for the early detection of environmental issues in Indonesia by leveraging social media data, particularly from Twitter. Environmental problems such as air pollution and deforestation pose serious risks to public health, biodiversity, and economic sustainability. However, traditional monitoring systems are often delayed or limited in coverage. To address this, we combined IndoBERT a pretrained language model for Indonesian for sentiment analysis and entity extraction, with Latent Dirichlet Allocation (LDA) for topic modeling. The dataset, collected using specific keywords related to pollution and deforestation, underwent a rigorous preprocessing pipeline before analysis. Results show that public sentiment is predominantly negative, reflecting strong concerns about air quality and illegal logging. LDA revealed coherent topic clusters, such as haze-related urban pollution and deforestation linked to mining and palm oil expansion. These findings highlight the potential of social media mining as a complementary tool for real-time environmental monitoring. The proposed framework provides actionable insights for policymakers, NGOs, and smart city platforms to detect and respond to emerging environmental threats more proactively.