Using Demographic Information to Reduce the New User Problem in Recommender Systems

Abstract: Recommender systems rely heavily on user data to make accurate rec- ommendations. This presents a problem for new users for whom no such data is available. This study investigated if this problem could be reduced by basing recommendations solely on user’s demographic in- formation. Experiments were conducted using a framework that em- ploys K-means clustering. To evaluate the framework, the MovieLens 100K dataset was applied to a set of experiments. While the results did not exhibit any correlation between ratings and demographic features in the MovieLens 100K dataset, it does not exclude that the framework is not effective on other datasets with more demographic features.