When the most "pertinent" document should not be retrieved - An analysis of the Swets model

Abstract Most automated information retrieval systems operate by relating a document to a request by means of a measure of pertinance, and then retrieving the most pertinent documents for their patrons. In this paper the consistency of this operating procedure with the well known Swets Model is examined. It is shown that accepting the assumptions made by Swets would result in the possible rejection of the most pertinent documents in favor of those that are less pertinent. This conclusion is a consequence of the normality assumptions of the model, while other distributions, such as the Poisson distribution, is consistent with the standard procedure. In the course of the development, the fundamentals of decision theory and signal detection theory are reviewed.