Issue No. 09 - Sept. (2012 vol. 24)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/TKDE.2011.118
Andreas Pashalidis , K.U.Leuven, Leuven and IBBT, ESAT/SCD-COSIC
Bart Preneel , K.U.Leuven, Leuven and IBBT, ESAT/SCD-COSIC
While personalization is key to increase the usability of online services, disclosing one's preferences is undesirable from a privacy perspective, because it enables profiling through the linkage of what may otherwise be unlinkable service invocations. This paper considers an easily implementable class of obfuscation strategies as a means to mitigate these risks, and examines its privacy/utility tradeoff. Our results are based on simulations that take place within a modular evaluation framework that can seamlessly accommodate real-world data. We conducted experiments with different simulated behaviors and using two preference populations, namely a population of maximally diverse preferences and one consisting of the movie preferences of some Netflix users. We measure utility in a way that is specific to the application of preference obfuscation. Privacy is measured in terms of unlinkability, with respect to two different adversaries. Our results show that reasonable privacy/utility tradeoffs require the disclosure of only small amounts of preference information.
Privacy, Servers, Bayesian methods, Clustering algorithms, Simulated annealing, Partitioning algorithms, Probability distribution, unlinkability, Preferences, obfuscation, privacy
B. Preneel and A. Pashalidis, "Evaluating Tag-Based Preference Obfuscation Systems," in IEEE Transactions on Knowledge & Data Engineering, vol. 24, no. , pp. 1613-1623, 2011.