Bridging the Urban Divide: Adaptive Cross-City Learning for Disaster Sentiment Understanding
New framework combines social media text with real-world movement patterns to reduce urban bias.
Deep Dive
A research team led by Zihui Ma and Runlong Yu developed an adaptive cross-city learning framework for disaster sentiment understanding. The model integrates mobility-informed behavioral signals with textual data from social media. Focusing on the 2025 Southern California wildfires, it achieved state-of-the-art performance by using city similarity-based data augmentation. This multimodal fusion improves both accuracy and fairness, revealing geographically diverse sentiment patterns often missed by text-only models.
Why It Matters
Enables more equitable disaster response by understanding underrepresented communities, moving beyond urban-centric AI biases.