Recommender services that are currently used by IPTV providers help customers to find suitable content according to their preferences and increase overall content sales. Such systems provide competitive advantage over other IPTV providers and improve the overall performance of the current systems by building up an overlay that increases content availability, prioritization and distribution that is based on users' interests. Current implementations are mostly centralized recommender service (CRS) where the information about the users' profiles is stored in a single server. This type of design poses a severe privacy hazard, since the users' profiles are fully under the control of the CRS and the users have to fully trust the CRS to keep their profiles private. In this paper, we present our approach to build a private centralized recommender service (PCRS) using collaborative filtering techniques and an agent based middleware for private recommendations (AMPR). The AMPR ensures user profile privacy in the recommendation process. We introduce two obfuscation algorithms embedded in the AMPR that protect users' profile privacy as well as preserve the aggregates in the dataset in order to maximize the usability of information for accurate recommendations. Using these algorithms provides the user complete control on the privacy of his personal profile. We also provide an IPTV network scenario that uses AMPR and its evaluations.