SENTIMENT ANALYSIS OF CPEC ON TWITTER/X: PUBLIC OPINION TRENDS AND PREDICTORS OF SUPPORT

Abstract

Purpose: The China–Pakistan Economic Corridor (CPEC), a central component of China’s Belt and Road Initiative (BRI), has generated extensive public debate across Pakistan and beyond. While earlier research has focused on economic, geopolitical, and media perspectives, limited empirical work has examined how public sentiment toward CPEC evolves in digital environments. This study analyzes public opinion trends regarding CPEC on Twitter/X using sentiment scoring and time-series analysis to identify patterns, fluctuations, and predictors of support.

Methods: The dataset consisted of 680,000 tweets collected via the Twitter Academic API between January 2020 and December 2023 using CPEC-related keywords. A mixed NLP approach was used, including VADER sentiment scoring and a fine-tuned BERT classifier. Sentiment polarity was aggregated over time and analyzed using ARIMA and Granger causality testing to identify event-driven trends. Predictive models, including logistic regression and random forest, examined linguistic and engagement-related predictors of sentiment.

Results: Findings show polarized sentiment distribution (positive 41%, negative 36%, neutral 23%) and event-driven fluctuations. Positive sentiment peaks aligned with government announcements and infrastructure milestones, while negative sentiment aligned with security incidents, debt concerns, and regional political tensions. Engagement-related features, including tweet visibility and verified status, predicted supportive sentiment, whereas negative sentiment was strongly associated with sovereignty and security discourse.

Conclusion: Public sentiment toward CPEC on Twitter/X is dynamic, polarized, and highly responsive to political and economic events. The results highlight the role of digital platforms in shaping public perception of national megaprojects and underscore the importance of transparent communication and inclusive public discourse. Future research should explore cross-platform analysis, multilingual sentiment models, and public engagement effects.