The challenge in empirical development of information retrieval systems at TREC is obtaining general conclusions from IR system responses to a sample of perhaps 50 topics, that is, 50 statements of information need. Treating such a sample as a simple random sample and using a univariate performance measure to describe each search result does not provide sufficient sensitivity for many purposes. In this paper, we contend that greater sensitivity can be gained through study of the natural language basis for the dependence of system performance on topic.
Proceedings Title: Proceedings of the 53rd Session of the International Statistical Institute
Conference Dates: August 22-29, 2001
Conference Location: Seoul, KO
Conference Title: International Statistical Institute
Pub Type: Conferences
information retrieval, multidimensional scaling, query formation, rank correlation, search engines, system evaluation