Recall at K measures the share of correctly identified relevant items in the top K recommendations out of the total number of relevant items in the dataset (without taking into account the position of the item in the list). Or in simpler terms; how many of all the existing relevant items could be successfully find in the top K.