We will be using one of the large MovieLens (link) data sets to build a recommender system. In addition to building a user based collaborative filter and SVD, we will attempt to build a recommender that treats different users differently. Building on the “harbingers of failure” model, we want to tease out if there are some viewers who are particularly hard or easy to satisfy. We will treat users that are in line with the average ratings differently from idiosyncratic users. Some users may like movies that are not as popular. We will try to accomplish this by taking a deeper dive into the previous ratings of the users and different qualities of the movies.