Reference Work Entry

Encyclopedia of Database Systems

pp 193-194

Average Precision at n

  • Nick CraswellAffiliated withMicrosoft Research Cambridge
  • , Stephen RobertsonAffiliated withMicrosoft Research Cambridge




Average Precision at n is a variant of Average Precision (AP) where only the top n ranked documents are considered (please see the entry on Average Precision for its definition). AP is already a top-heavy measure, but has a recall component because it is normalized according to R, the number of relevant documents for a query. In AP@n there are a number of options for normalization, for example, normalize by n or normalize by min(n,R).

Key Points

The well-known measure Average Precision has a number of lesser-known variants, used in TREC [3] and elsewhere. Before and during TREC-1, it was usual to calculate an 11-point interpolated Precision-Recall curve, and take the average of these 11 precision values, giving an “interpolated AP.” In TREC-2 and beyond, the modern non-interpolated AP was introduced. It calculates precision at each relevant document.

A number of other AP variants arise in a precision-oriented setting, where it is possible ...

This is an excerpt from the content