Movie recommender systems attempt to find movies which are of interest for their users. However, as new movies are added, and new users join movie recommendation services, the problem of recommending suitable items becomes increasingly harder. In this paper, we present a simple way of using a priori movie data in order to improve the accuracy of collaborative filtering recommender systems. The approach decreases the sparsity of the rating matrix by inferring personal ratings on tags assigned to movies. The new tag ratings are used to find which movies to recommend. Experiments performed on data from the movie recommendation community Moviepilot show a positive effect on the quality of recommended items.