Adaptivity through Unobstrusive Learning

In this paper, we present an approach for learning interest profiles implicitly from positive user observations only. This approach eliminates the need to p rompt users for ratings, or to somewhat artificially infer negative evidences, which arises when traditional learning algorithms are used. We developed a methodology for learning explicit user profiles and recommending interesting objects. This highly dynamic process, which calculates the personalized recommendations in real-time, has been d eployed in ELFI, a web-based system that provides information about research grants and is used by more than 1000 users in German research organizations who monitor and/or advise on extra-mural funding opportunities.