Discovering relevant publications for researchers is a non- trivial task. Recommender systems can reduce the effort required to find relevant publications. We suggest using a visualization- and user-centered interaction model to achieve both a more trusted recommender system and a system to understand a whole research field. In a graph-based visualization papers are aligned with their keywords according to the relevance of the keywords. Relevance is determined using text-mining approaches. By letting the user control relevance thresholds for individ- ual keywords we have designed a recommender system that scores high in accuracy, trust, and usability in a user study, while at the same time providing additional information about the field as a whole. As a result, the inherent trust issues con- ventional recommendation systems have seem to be less significant when using our solution.