Most existing information filtering systems promote items that match a user’s previous choices or those that are popular among other similar users. This results in recommendations that are highly similar to the information users are already exposed to, resulting in their isolation inside familiar but insulated information silos. In this context, we propose to develop a recommendation framework with a goal of improving information diversity. We focus on the problem of political content recommendation without any assumption about the availability of labels regarding the bias of users or content producers. By exploiting social network signals, we propose to first estimate ideological positions for both users and the information items they share. Based on these positions, we then generate diversified personalized recommendations using a modified random-walk based recommendation algorithm. With experimental evaluations on large datasets of twitter discussions, we show that our method based on random walks with erasure is able to generate more diverse recommendations. This research addresses a general problem and can be extended to recommendations in other domains, as we show with experiments on open benchmark datasets.