Digital platforms have democratized news production and facilitated access to a wider range of sources. Many news providers are driven more by short-term financial goals and/or political interests, while readers are driven, at least in part, by the goal of making sense of the world through access to reliable information. This misalignment is exacerbated by the fact that social networks, which are also driven by profit, often act as intermediaries between sources and readers. The selection of news that social network users read is largely determined by recommendation algorithms that prioritize articles that are assumed to match users' preferences, regardless of quality and veracity. Existing work [2] shows that recommenders lead to polarization in the long run, a phenomenon that is detrimental to public debate and ultimately to democracy. We present work in progress from the BOOM project 1 , which focuses on understanding and characterizing polarization, and then proposing recommendation strategies that reduce it. The focus of the presented work is the automatic and fine-grained characterization of the stance of news articles using Twitter as a source of political knowledge.