A Hypergraph-Based Defense Framework Against Relational Inference in Mobile Crowdsensing

Abstract

Repeated co-occurrences in worker trajectory data can expose latent relationships in mobile crowdsensing. Existing methods mainly focus on protecting individual or dyadic relationships, which inadequately capture higher-order relationships between trajectories. To address this problem, we propose a hypergraph-based defense framework (HGDF) against relational inference in mobile crowdsensing. Specifically, we first mine repeated spatiotemporal co-occurrence events to identify dyadic and higher-order relationships. Then, we model dyadic relationships with a graph and higher-order relationships with a hypergraph. Next, HGDF maps each exposed relation to the trajectory records that form its co-occurrence evidence. Finally, the remaining spatiotemporal obfuscation options are evaluated by comprehensively scoring evidence reduction, newly exposed risks, and distortion cost. The highest scoring candidates are iteratively applied until all protected relationships fall below the inference thresholds. Theoretical analysis shows that HGDF satisfies the specified exposure-threshold constraints for protected dyadic and higher-order relationships. We evaluate HGDF on the GeoLife dataset and show that it suppresses both dyadic and higher-order co-occurrence more effectively than representative baselines while modifying fewer trajectory records.