Public opinion is essential for assessing whether mega infrastructure projects (MIPs) meet societal expectations. Social media, as a participatory information source, facilitates the analysis of MIP-related opinions. However, previous studies have overlooked embedding-based topic modeling and the link between topics and sentiments. To address this, the study proposes a spatiotemporal framework to dynamically extract topics and analyze public sentiment from unstructured online data.
An embedding-based topic extraction model automatically identifies topics across regions and stages, while a pre-trained sentiment analysis model fine-tuned from Paddle Natural Language Processing evaluates public attitudes toward specific topics. The framework is validated on the Guangzhou-Shenzhen-Hong Kong High-Speed Railway project using 26,891 posts and 55,471 comments over nine years on social media.
Results reveal variations in topics and sentiment across the construction, opening and operational stages among regions. Positive sentiments dominated (78.5%) across 121 topics, while negative sentiments stemmed from delays, cost overruns, high ticket prices and limited transparency.
This study provides actionable strategies for managing large-scale MIPs, including establishing cross-regional coordination committees, enhancing target management and prioritizing public participation.
This study addresses a gap in literature by integrating embedding-based topic modeling and sentiment analysis to explore the association between topics and sentiments. It offers a data-driven approach for dynamic public opinion analysis, contributing to improved management practices for MIPs.
