The reliability of metrics based on graded relevance

Abstract

This paper compares 14 metrics designed for information retrieval evaluation with graded relevance, together with 10 traditional metrics based on binary relevance, in terms of reliability and resemblance of system rankings. More specifically, we use two test collections with submitted runs from the Chinese IR and English IR tasks in the NTCIR-3 CLIR track to examine the metrics using methods proposed by Buckley/Voorhees and Voorhees/Buckley as well as Kendall's rank correlation. Our results show that AnDCGl and nDCGl ((Average) Normalised Discounted Cumulative Gain at Document cut-off l) are good metrics, provided that l is large. However, if one wants to avoid the parameter l altogether, or if one requires a metric that closely resembles TREC Average Precision, then Q-measure appears to be the best choice. © Springer-Verlag Berlin Heidelberg 2005.