Physical Trajectory Inference Attack and Defense in Decentralized POI Recommendation

As an indispensable personalized service within Location-Based SocialNetworks (LBSNs), the Point-of-Interest (POI) recommendation aims to assistindividuals in discovering attractive and engaging places. However, theaccurate recommendation capability relies on the powerful server collecting avast amount of users' historical check-in data, posing significant risks ofprivacy breaches. Although several collaborative learning (CL) frameworks forPOI recommendation enhance recommendation resilience and allow users to keeppersonal data on-device, they still share personal knowledge to improverecommendation performance, thus leaving vulnerabilities for potentialattackers. Given this, we design a new Physical Trajectory Inference Attack(PTIA) to expose users' historical trajectories. Specifically, for each user,we identify the set of interacted POIs by analyzing the aggregated informationfrom the target POIs and their correlated POIs. We evaluate the effectivenessof PTIA on two real-world datasets across two types of decentralized CLframeworks for POI recommendation. Empirical results demonstrate that PTIAposes a significant threat to users' historical trajectories. Furthermore,Local Differential Privacy (LDP), the traditional privacy-preserving method forCL frameworks, has also been proven ineffective against PTIA. In light of this,we propose a novel defense mechanism (AGD) against PTIA based on an adversarialgame to eliminate sensitive POIs and their information in correlated POIs.After conducting intensive experiments, AGD has been proven precise andpractical, with minimal impact on recommendation performance.