Lior Kirsch

Lets say that we are in the movies recommandation business and instead of just providing yes/no recommendation we want to provide a 5 values scoring system: [“don’t bother”, “bad but watchable”, “nice”, “very nice”, “Shawshank Redemption” ]. We have this ordering in mind that movies that are “nice” are better than “don’t bother” but are worse than “very nice” and also from “Shawshank Redemption” which has a category all of its own.

AUC - Area under the curve, usually refers to the area under the ROC curves. It is common practice in machine learning and in statistics to plot the ROC curves and from there gain information about the behaviour of the classifier. I would like to talk about AUC but not in the classical context of ROC curves. Another less known interpretation, is that AUC is the probability that a positive sample will receive a score higher than a negative sample.