Thanks Christoph. Are these numbers for mllib als implicit and explicit feedback on movielens/netflix datasets documented on JIRA ? On Sep 19, 2014 1:16 PM, "Christoph Sawade" < christoph.saw...@googlemail.com> wrote:
> Hey Deb, > > NDCG is the "Normalized Discounted Cumulative Gain" [1]. Another popular > measure is "Expected Reciprocal Rank" (ERR) [2]; it is based on a > probabilistic user model, where the user scans the presented list of search > results or recommendations and chooses the first that is sufficiently > relevant. ERR is the expectation of the reciprocal rank under this model. > Precision@k is the precision at a cut-off k. Often this is used as an > overage over the first n k's; a starting point might by [3]. > > Hope that helps. > > Cheers, Christoph > > [1] K. Järvelin and J. Kekäläinen. Cumulated gain-based evaluation of IR > techniques. ACM Transactions on Information Systems, 20(4):422–446, 2002. > [2] O. Chapelle, D. Metzler, Y. Zhang, and P. Grinspan. Expected > reciprocal rank for graded rel- evance. In Proceeding of the Conference > on Information and Knowledge Management, 2009. > [3] http://en.wikipedia.org/wiki/IR_evaluation >
