Location-based applications offer increasingly personalized services to mobile users. Incorporating temporal and demographic information can further improve service quality. However, sharing such information carries the risk of leaking private data, including a user’s identity or further personal attributes. Differential Privacy (DP) is a widely accepted privacy notion to protect user data in this context. However, DP does not account for adversarial background knowledge, which can undermine privacy through context linking attacks. To design resilient privacy mechanisms, a systematic analysis is required to determine which pieces of background information pose the highest risk. 
In this work, we investigate whether knowing the privacy mechanism and semantic information can break DP and enable an adversary to reconstruct a user’s location. 
We evaluate which types of background knowledge contribute most to attack success by designing a series of attacks with increasing access to semantic context, such as points of interest (POIs), mobility statistics, demographic data, and privacy parameters. We conduct an extensive evaluation on two large datasets. Our results show that knowledge of POIs and typical mobility patterns, especially when combined with the privacy parameter, substantially increases attack success, particularly in rural areas and for certain demographic groups.