Given the rapid growth of participatory media content such as blogs, there is a need to design personalized recommender systems to recommend only useful content to users. We believe that in addition to producing useful recommendations, certain insights from media research such as simplification and opinion diversity in recommendations should form the foundations of such recommender systems, so that the behavior of the systems can be understood more closely, and modified if necessary. We propose and evaluate such a system based on a Bayesian user-model. We use the underlying social network of blog authors and readers to model the preference features for individual users. The initial results of our proposed solution are encouraging, and set the agenda for future research.