One-Class Matrix Factorization

The second topic, we are investigating is binary classification with matrix factorization models where only observations of one class are available. But in contrast to the problem settings we have discussed so far (see chapter 3), we now deal with a binary classification problem where the classes are more or less balanced. Context-aware ranking like item recommendation (see chapter 6) differs from this because it is a ranking task. Even when seeing it as a binary classification task, the problem differs substantially because the classes are typically hugely imbalanced: e.g. a customer buys much less books than he does not buy, a user listens to much less songs than he never listens to, etc.