Private recommendation service for IPTV systems: Protecting user profile privacy

IPTV providers employ third party content recommendation service to help end users find personalized content, and at the same time increase content sales and gain competitive advantage over other IPTV providers. However current implementations of recommendation services are mostly centralized where all the information about the users' profiles is stored on a dedicated server. A common fear among users is that their user profile data being misused by recommendation provider. Also sharing profile data makes the end users vulnerable to attacks like insider attacks, where an employee of the recommendation service may compromise the confidentiality and integrity of their profiles. For these reasons, privacy aware users intentionally decline to use recommendation or even provide inaccurate or wrong information because they consider it as untrusted service. On the other hand, to build an accurate recommendation model the user must reveal information that is typically considered private such as watching history, previous buying behaviour, content ratings, etc. Further privacy concerns arise when the user data are stored in countries that have privacy laws different from the country where the service is consumed. This poses a severe privacy hazard, since the users profiles are fully under the control of recommendation provider and stored in locations that are not legally bound to ensure the privacy of its users. Due to different legal structures that relate to data privacy laws in different legal jurisdictions maintaining user profile privacy is not a trivial solution. Regardless of the official legal framework requirements, when outsourcing users' profiles the private data should be kept safe when it is in the possession of any third party service. In this paper we introduce a private recommendation method using collaborative filtering techniques. The method preserves the privacy of its users when using the system and allows sharing data among different users in the network. We also introduce two obfuscations algorithms that protect users profile privacy as well as preserve the aggregates in the dataset to maximize the utility of information to provide accurate recommendations. Using these algorithms provides the privacy of users personal profiles.