Privacy Aware Recommender Service using Multi-agent Middleware - An IPTV Network Scenario

IPTV service providers are starting to realize the significant value of recommender services in attracting and satisfying customers as they offer added values e.g. by delivering suitable personalized contents according to customers personal interests in a seamless way, increase content sales and gain competitive advantage over other competitors. However the current implementations of recommender services are mostly centralized combined with collecting data from multiple users that cover personal preferences about different contents they watched or purchased. These profiles are stored at third-party providers that might be operating under different legal jurisdictions related to data privacy laws rather than the ones applied where the service is consumed. From privacy perspective, so far they are all based on either a trusted third party model or on some generalization model. In this work, we address the issue of maintaining users' privacy when using third-party recommender services and introduce a framework for Private Recommender Service (PRS) based on Enhanced Middleware for Collaborative Privacy (EMCP) running at user side. In our framework, PRS uses platform for privacy preferences (P3P) policies for specifying their data usage practices. While EMCP allows the users to use P3P policies exchange language (APPEL) for specifying their privacy preferences for the data extracted from their profiles. Moreover, EMCP executes a two-stage concealment process on the extracted data which utilize trust mechanism to augment the recommendation's accuracy and privacy. In such case, the users have a complete control over the privacy level of their profiles and they can submit their preferences in an obfuscated form without revealing any information about their data, the further computation of recommendation proceeds over the obfuscated data using secure multi-party computation protocol. We also provide an IPTV network scenario and experimentation results. Our results and analysis shows that our two-stage concealment process not only protect the users' privacy, but also can maintain the recommendation accuracy.