A privacy‐preserving framework for location recommendation using decentralized collaborative machine learning

Abstract

The nowadays ubiquitous location‐aware mobile devices have contributed to the rapid growth of individual‐level location data. Such data are usually collected by location‐based service platforms as training data to improve their predictive models' performance, but the collection of such data may raise public concerns about privacy issues. In this study, we introduce a privacy‐preserving location recommendation framework based on a decentralized collaborative machine learning approach: federated learning. Compared with traditional centralized learning frameworks, we keep users' data on their own devices and train the model locally so that their data remain private. The local model parameters are aggregated and updated through secure multiple‐party computation to achieve collaborative learning among users while preserving privacy. Our framework also integrates information about transportation infrastructure, place safety, and flow‐based spatial interaction to further improve recommendation accuracy. We further design two attack cases to examine the privacy protection effectiveness and robustness of the framework. The results show that our framework achieves a better balance on the privacy–utility trade‐off compared with traditional centralized learning methods. The results and ensuing discussion offer new insights into privacy‐preserving geospatial artificial intelligence and promote geoprivacy in location‐based services.

Information Security and Privacy in Railway Transportation: A Systematic Review

“…Additionally, the simulations report an average delay in mitigating attacks (under normal operation ranges) from 0.2 to 0.6 s. A secure integration with external systems is considered in future work. Furthermore, based on a decentralised collaborative ML approach, [ 40 ] introduces a privacy-preserving location recommendation framework that keeps users’ data on their devices. Although the framework has some limitations (e.g., long time to train the model, high need of computing resources…), the offered method could be considered as an important step towards the implementation of privacy-preserving AI technologies.…”

“…Thus, accessing public wireless networks or sharing private information (e.g., location) might also be a source of harmful cyber incidents. Relatedly, the development of models to guarantee passengers’ privacy protection will improve the resilience against cyberattacks [ 40 ]. Although travellers’ data can be used by railway companies to improve their safety, security and QoS, it can also lead to complex social and ethical challenges in terms of accountability and transparency [ 37 ].…”

A comprehensive survey on online social networks security and privacy issues: Threats, machine learning‐based solutions, and open challenges

“…Alyousef et al 145 proposed an intelligent location privacy scheme using deep learning and a set of dummy locations where attackers cannot distinguish the real location of a user. Rao et al 146 also proposed an ML‐based approach to store users' location information securely and they also proved the effectiveness of the approach by simulating two location privacy attacks.…”

Decentralized Collaborative Learning Framework for Next POI Recommendation

“…To tackle this issue, we ask the server to perform neighbor identification for each user on the server side. Compared with federated frameworks that also cluster similar users to facilitate better model personalization [36,42], DCLR avoids collecting users' raw data which is highly sensitive, and instead leverages implicit user preference indicators to support accurate yet privacy-aware neighbor identification. Furthermore, after the self-supervised pretraining task, the server is only responsible for identifying neighbors for users, and is released from the hefty role of aggregating all local models throughout the training cycle which is computationally expensive.…”

“…Moreover, all users share the same global model, which ignores the dynamics and diversity of users' spatial activities and interests, leading to suboptimal performance. Although some approaches [36,42] can cluster clients to provide group-based personalized models, this commonly involves the transmission of sensitive user attribute information. To conclude, we are still in pursuit of a decentralized POI recommendation paradigm that requires minimal engagements and resources from the central party, and can collaboratively learn personalized models amid highly sparse interaction data at the individual device level.…”