Content personalization is identified as a key technology for enabling ubiquitous access to social media. Recommender systems implement media personalization, by suggesting relevant content and helping users in addressing the “information overload” problem. In this paper, our aim is to improve personalization by increasing the accuracy of recommendations. We propose a novel method, called Content Relationships Matrix Factorization (CRMF), which exploits additional information in the form of content relationships that express relevance between items. We model content relationships based on affinity graphs and use them in the context of matrix-factorization, which are currently the state-of-the-art prediction models for recommender systems. In our experimental evaluation with a real data set, we demonstrate the accuracy improvement of CRMF compared to matrix factorization models that do not take into account content relationships. Our experimental results show that CRMF compares favorably to the baseline method, demonstrating the usefulness of considering content relationships.