Pricavy-Preserving Aspects for Data Mining in ICT Services

The steady adoption of systems for profiling users behavior, collecting and critically interpreting as much information as possible about likes and dislikes, interests and habits of Internet residents and generic services consumers have rapidly become some of the hottest keywords within networking research community. Indeed, mining information about users behavior is an advantage for both service providers and service customers: on one side, providers can improve their revenues by focusing on the most successful features of their services, while on the other side, users can enjoy services which reflect closer their specific needs. There are many examples of user profiling applications. Internet television platforms have earned a tremendous success in recent years, and one of the key factor of their economical profits has been undoubtedly the adoption of p2p architectures. The flourishing of plenty of channels with well defined thematic strands has contributed to polarizing the users’ choice towards those offerings what better reflects their likes. In parallel, several public utility systems (such as the electricity, the gas, and the water distribution systems) are deploying smart meters for improving the management of the distribution network and offering better services to the users. Also very generic and largely used internet services, such as DNS (Domain Name System) service, can be exploited for monitoring user data related to the most visited web sites. Although tempting, the scenario where users’ habits are constantly tracked hides privacy violations and annoyance: much can be learnt about the lifestyle and behaviors of the customers monitoring their electricity consumptions, music and movie downloads from the Internet, most frequently visited sites, and so on. To preserve the advantages of profiling while confining the drawbacks, the goal of this thesis is defining some technical solutions which allow to mine users’ data in a privacy preserving manner. We therefore put forth some architectural alternatives to release aggregate information about the data provided by multiple users, without leaking information about individual data: so we talk about privacy preserving Data Mining tasks. The tools on the basis of our approach are Secure Multi-Party Computation and Secret Sharing, which allow to perform the vector addition operations required by the majority of the Data Mining algorithms without revealing user private data.