Geo-stance-aware recommender for climate action via spatio-temporal multilayer networks

Abstract

This study introduces a geo-stance-aware recommendation framework that links geolocated social media climate discussions with real-world disaster events to reduce information silos and ideological polarization. The framework is designed to address location-specific climate risks (e.g., rising sea levels, droughts, wildfires) by connecting disparate user communities through shared topics and nearby events. We construct a heterogeneous multilayer network of users, topics, and disaster events, with edges weighted by sentiment alignment, spatial proximity, and temporal recency. Our network analysis, featuring Stance-Aware Eigenvector Centrality (SAEC) and a sentiment-temporal enhancement of the Louvain community detection algorithm, achieves a modularity of 0.861. Experiments demonstrate that the system identifies bridging topics (those that connect diverse ideological groups) and key disaster events, effectively breaking down echo chambers and improving the spread of balanced climate information. The resulting recommendation engine delivers actionable, location-aware content that accounts for user stance, sentiment, and proximity to events. These targeted recommendations have the potential to catalyze local climate resilience interventions and foster community preparedness and adaptation.