Improving Vaccine Stance Detection by Combining Online and Offline Data

Both online and offline information matters for understanding vaccine hesitancy: The paper demonstrates that vaccine stance cannot be accurately predicted from social media alone—offline factors (such as demographic characteristics, local community influences, and real-world experiences) are equally important complements to online discourse analysis.

Real-world context is critical for responsible AI applications in public health: By identifying that offline attributes significantly influence vaccine decisions, this research highlights why AI systems designed to detect vaccine hesitancy must account for factors beyond what appears in online data, preventing incomplete or biased assessments of public health attitudes.

Explainable AI helps identify which factors actually drive vaccine decisions: The paper uses explainable AI methods to pinpoint the specific online and offline attributes that influence vaccine stance, rather than treating prediction as a "black box"—this transparency is essential for policymakers and health communicators to understand what's actually shaping public opinion.

Integrating diverse data sources improves accuracy and fairness: By combining online and offline data, the model achieves better predictions of vaccine stance, suggesting that responsible AI systems should draw from multiple information sources to avoid over-relying on any single perspective or demographic group.