With the unprecedented growth of user-generated content produced on microblogging platforms, finding interesting content for a given user has become a major issue. However due to the intrinsic properties of microblogging systems, such as the vol-umetry, the short lifetime of posts and the sparsity of interactions between users and content, recommender systems cannot rely on traditional methods, such as collaborative filtering matrix factorization. After a thorough study of a large Twitter dataset, we present a propagation model which relies on homophily to propose post recommendations. Our approach relies on the construction of a similarity graph based on retweet behaviors on top of the Twitter graph. Finally we conduct experiments on our real dataset to demonstrate the quality and scalability of our method.