Learning to rank using Lenskit -

i using lenskit framework learning rank problem. have issue have never seen reporting.
when compare performance of funksvd , popularity baseline on whole universe of items, popularity baseline outperforms funksvd. think cased fact popular items rated more often, while not mean users popular items.
limited number of candidate items evaluation n popular ones. result, performance of funksvd increased, since algorithm has less opportunity make mistake. funksvd outperformed popularity. however, have issue. random baseline beats both funksvd , popularity! guess that's because chance make mistake low, conduced. there way make funksvd perform better popularity , random? doing wrong? problem in framework? has encountered problem?

how measuring performance? key thing make significant difference.

also, funksvd has number of parameters must tuned in order have performance. default values may not data set.

diagnosing going wrong specific experiment seems better handled in lenskit discussion fora (mailing list or gitter room).