I was wondering if there is a source, where given some train data and test data, the test statistics of evaluation of Recommender Systems are also provided. For example, given two files train.dat and test.dat, where the data have already been split into a training and a test set which contain user_id, item_id and ratings (just like in grouplens dataset) and in the end some answer for precision or recall or map@k test is provided for the performance of a nonpersonalized recommender system (like the top rated items-most viewed) or any other recommender system.
Thank you in advance,
Regards,
Marios
The fastest way to find out how known recommenders performance on the new dataset is to grab existing implementation and run it. If You have prepared split - it's OK, but make sure that splitting strategy is adequate what You are trying to measure.
Helpful project for evaluation: - librec - project Rival - MediaLens
And make sure, the right measure is used for the right purpose. In the question the rating prediction example is given and then the ranking evaluation measures are listed.