SAFAIR AI Contest
from 1st March to 31st May
All accepted publications from SPARTA partners under its funding.
MobHide: App-level Runtime Data Anonymization on Mobile
Davide Caputo, Luca Verderam, Alessio MerloAbstract
Developers of mobile apps gather a lot of user’s personal information at runtime by exploiting third-party analytics libraries, without keeping the owner (i.e., the user) of such information in the loop. We argue that this is somehow paradoxical. To overcome this limitation, in this paper, we discuss a methodology (i.e., MobHide), allowing the user to choose a different privacy level for each app installed on her device. According to the user’s preferences, MobHide anonymizes the data collected by the analytics libraries before sending them to the app developers, through a fruitful combination of data anonymization techniques. More in detail, the methodology enables to i) analyze all the network traffic generated by the invocation of analytics libraries, ii) anonymize the personal and device data using a generalization technique, and the events related to the user’s behavior by exploiting local differential privacy, and iii) send the anonymized data to the developers.
We empirically assessed the viability of the approach on Android, by implementing the methodology as an Android app, i.e., HideDroid, that relies on the VPN service provided by Google to intercept all network requests. Our preliminary experiments - carried out on a real app (i.e., Duolingo) - are promising, and suggest that runtime data anonymization on mobile is feasible nowadays, as it negligibly impacts the app performance.