Precision at K

Also known as · Precision@K

The share of the top K recommended items that are relevant: the number of relevant items in the top K divided by KK. Relevant usually means rated 4 or above.

When to use

The homepage metric for a recommender: of the 5 shows shown, how many will the user enjoy. Worked example in the lecture: top 5 with 4 hits is 4/5=0.804/5 = 0.80. It is measured by hiding some known ratings as a test set. It matters more for a homepage, Recall at K more for search.

See DS Lec 6.

Appears in

No references yet.