Social media sentiment during infrastructure protests: a natural language processing approach to citizen engagement in China

Social discourse concerning public infrastructure projects in China has been highly controversial, and social media platforms have emerged as a significant venue of public expression. The paper takes an approach based on Natural Language Processing (NLP) to examine the sentiment behind the long-running opposition to Chinese infrastructure projects observed across key social media platforms (Weibo, WeChat, etc.) between 2022 and 2024. There was a collection of 1.2 million posts built on Weibo, WeChat public accounts, and online forums that were preprocessed through tokenization, stop-words, and word segmentation of Chinese characters. The Bidirectional Encoder Representations from Transformers (BERT) model was fine-tuned to be able to classify the sentiment of tweets with an overall accuracy of 91.3%. A trend analysis suggests that sentiment peaks in negativity coincide with key protest actions and official statements by the government, whereas peaks in positivity correlate with the messages about long-term advantages of development. Topic models indicated that citizen concerns are grouped in terms of displacement, environmental impact, and project planning transparency. Its findings suggest that NLP offers the promise of real-time understanding of the flow of ideas and opinion among citizens, which policymakers could use to develop strategies to engage with a citizenry and understand their issues and concerns to avoid future conflicts. This study is relevant to the fields of computational social science and government management by providing an example of how sentiment analysis may be used to track civic engagement in the context of a particularly politically delicate infrastructure project.