Table 3:

For 100 test patients, their first 20 orders are used as query items to get 10 recommended orders from each respective method. These top 10 recommendations are compared against the actual next set of up to 10 orders for each patient. Recall, precision, and F1-score is calculated for each method and averaged across all test patients.

Method	Recall	Precision	F1-Score	Method Description
Random	0.3%	0.3%	0.3%	Items randomly recommended from available catalog
BaselineFreq	14.4%	13.2%	13.5%	General “best seller” list, recommending overall most common orders
ItemAssociation	19.8%	18.2%	18.7%	Items ranked based on conditionalFreq(B|A) ∼ P(B|A)
NextDay	30.1%	27.8%	28.4%	Same as above, but uses nABday (only counts co-occurrences <1 day)
NextHour	23.8%	21.7%	22.2%	Same as above, but uses nABhour