Recommendation Evaluation: Precision@k

btibertbtibert Member, University Professor Posts: 146 Guru
In prep for my upcoming lecture, it is pretty straightforward how to demonstrate/discuss recommender systems via RM, though it's a bit unintuitive that we have to hand-enter item/user identification, but it's not the end of the world.

Having said that, in browsing the web for examples to provide, I noticed that there are times references to precision@k, but the only metrics that we get back as part of the performance operator are RMSE, MAE, and NMAE.  

Is there an operator that is available to do this and I am just missing it, which is often the case on my end.

Best Answer

  • Options
    btibertbtibert Member, University Professor Posts: 146 Guru
    Solution Accepted
    As is always the case, I was thinking about the problem incorrectly.  Obviously this only applies to item-based, and above, I am doing rating-based.  My mental model was more in line with what we do for classification or regression metrics, where the items are selectable.  

    This is a non-issue.


  • Options
    lionelderkrikorlionelderkrikor Moderator, RapidMiner Certified Analyst, Member Posts: 1,195 Unicorn
    Hi @btibert,

    Just in case, there is an extension (to install from the marketplace) called "Recommender" dedicated to recommender systems (...logic  :) ).
    In this extension, you have especially the operator Performance(Item Recommendation) which allows to calculate the precision@k...

    Hope this will help in the future...


Sign In or Register to comment.