Topic-sentiment analysis of citizen environmental complaints in China: Using a Stacking-BERT model.

Environmental complaints serve as a crucial means for citizens to participate in environmental regulation, providing precise insights for real-time identification of pollution issues and understanding public environmental concerns. This study analyzes 102,782 environmental complaint texts from China's e-government platform between 2016 and 2022, firstly employing an ensemble machine learning model (Stacking-BERT) to examine the hot topics and sentiment characteristics of these environmental complaints. The results indicate that: (1) The number of environmental complaints exhibits an "M-shaped" fluctuation, with the proportion of complaints related to noise, waste, and radiation continuously rising. The sentiment orientation of these complaints has shifted from predominantly negative (2016-2020) to more neutral (2021-2022), indicating a positive trend. (2) Complaints regarding air, water, land, noise, and waste demonstrate significant seasonal cyclical fluctuations, characterized by a homogenized pattern. (3) Approximately 70.41% of environmental complaints express negative emotions. Noise complaints are the dominant topic of negative emotions, while air and radiation complaints are the core themes of extreme negative emotions. (4) There is a high overlap between major complaint regions and hotspots of negative emotions. The dual hotspot areas (Guangdong, Hebei, Shandong, Henan) are identified as critical regions requiring urgent attention to resolve environmental complaints.