In the recommendation system space, it's called a sparsity problem (and/or a cold-start problem). You basically need to continuously feed it more data about your tastes for it to start doing a good job.