Average Precision Histogram

Definition

The average precision histogram plots the performance of a single run produced by an information retrieval system on a per-query basis. Each data point on the abscissa represents a query used in the evaluation process. The corresponding points on the ordinate measures the difference in this run’s performance on the given topic relative to the median average precision of other runs on that topic.

Key Points

Average precision histograms are often used to illustrate the performance differences that an information retrieval system may exhibit across different queries in an evaluation set in relation to other systems or runs that are evaluated on the same set of queries. These histograms can be used to identify the queries in an evaluation set which pose the most difficulty for information retrieval systems. Various tracks in the NIST Text Retrieval Conference (TREC) have used average precision histograms in the post-competition analysis of submitted runs [1]. An example average...