Visualization and User Control of Recommender Systems
Department of Computer Science
University of Saskatchewan, Canada
Abstract. Recommender systems are one of the “hottest” areas for companies that offer items to end users and have a wide scope of applications, from commercial, through entertainment to educational. Most recommender systems are based on implicit or explicit models of user behavior, including user ratings, click streams or navigation paths, or complex knowledge structures derived through reasoning from user input data. They deploy various personalization algorithms to adapt the recommendation to the user’s interest and context. The current recommendation algorithms have advanced to a stage where they offer very high accuracy in predicting which items the user will like. However, the resulting recommendations are not always accepted well by the users. Sometimes users prefer not to share their data with the system because of privacy concerns, or do not trust the system’s recommendations, especially if they do not understand how they were generated. To gain the user’s trust, it is important to explain the main principles or mechanism that the recommender system uses; however an explanation would increase the cognitive load of the user and therefore would not be welcome. A visualization of the mechanism or its main principles may be helpful here, since “a picture is worth a thousand words” and it can save users’ time and efforts by showing an intuitive and understandable representation of the recommendation mechanism. This talk will give an overview of some of the existing approaches (both from the speaker’s own research lab and from other authors) for visualizing recommendation mechanisms and eventually allowing users to control these mechanisms. Starting with work from the area of open/scrutable learner models in the area of intelligent tutoring systems, through approaches for explaining recommendations to approaches visualizing aspects of collaborative, hybrid and social recommenders, the talk will focus on approaches for visualizing social recommendation in streams of social sites updates.