# [2019.10.15] Cold-start for Users I've never solved a user cold-start problem before. I mean, I worked on an onboarding task. What to show to a new user for demonstrating the most popular but also most diverse items. I solved a problem of rating elicitation. I've never solved a problem of actual transfer learning. When you have piles of data about users based on their other activity, not directly related to your recommender system in focus. Usually, you don't have such data and don't work in such a scenario. First of all, it's difficult since frequently there are quite many users. So you can't easily calculate a full similarity matrix for them. That problem becomes intractable. And challenging!:)