Adaptive cloaking for contextual privacy in spatial crowdsourcing applications

Abstract

Spatial crowdsourcing systems have revolutionized mobility-driven applications by enabling users to contribute geolocated data in real time. However, this pervasive data collection exposes individuals to serious privacy risks, such as location tracking, profile inference, and behavioral disclosure. Traditional anonymization methods (including k-anonymity, dummy generation, and differential privacy) often fail to reflect the subjective, contextual, and semantic dimensions of user privacy. They lack the flexibility to adapt to user-specific needs, the ability to reason about the sensitivity of places, and the means to preserve the usability of anonymized data. In this paper, we introduce an adaptive cloaking framework for contextual location privacy called SPARC+. Our model integrates user-defined privacy rules, semantic metadata, and contextual signals to infer the privacy sensitivity of each location in real time. Then, it dynamically selects the most appropriate obfuscation strategy from a predefined set, including semantic substitution, dissimilarity, decoy injection, and temporal cloaking. Each strategy produces a personalized anonymization zone satisfying key privacy guarantees such as $$\theta $$-location indistinguishability, $$\beta $$-semantic diversity, and $$\tau $$-distributional protection. To guide the selection of dummy locations, SPARC+ uses a risk-weighted utility function that balances spatial dispersion with behavioral plausibility. Our probabilistic selection process is based on the exponential mechanism to ensure differential privacy. We validate our approach through a comprehensive experimental evaluation using real mobility traces from the Gowalla dataset and a synthetic dataset of simulated users, covering more than 100 individuals. The results confirm that SPARC+ effectively reduces the probability of re-identification while maintaining high semantic dispersion and geographic plausibility. Compared to state-of-the-art methods, our framework offers a more interpretable, context-aware, and user-controllable approach to location privacy, making it suitable for deployment in real-world spatial crowdsourcing platforms.