From books to movies to people to follow on Twitter, recommender systems carve the deluge of information on the internet into a more personalized flow, thus improving the performance of E-commerce, web, and social applications. It is no great surprise, given the success of Amazon-monetizing recommendations and the Netflix Prize, that any discussion of personalization or data-theoretic prediction would involve a recommender. What is surprising is how simple recommenders are to implement yet how susceptible they are to the vagaries of sparse data and over-fitting.

Consider a non-algorithmic approach to eliciting recommendations: one of the easiest ways to garner a recommendation is to look at the preferences of someone we trust. We are implicitly comparing our preferences to theirs, and the more similarities you share, the more likely you are to discover novel, shared preferences. However, everyone is unique, and our preferences exist across a variety of categories and domains...