We compared two different interfaces to the InQuery IR system with respect to their support for the TREC-9 Interactive Track Question-Answering task. One interface presented search results as a ranked list of document titles (displayed ten at one time), with the text of one document (the first, or any selected one) displayed in a scrollable window. The other presented search results as a ranked series of scrollable windows of the texts of the retrieved documents, displayed six documents at a time, each document display beginning at the system-computed "best passage". Our hypotheses were that: multiple-text, best passage display would have an overall advantage for question answering; single-text, multiple title display would have an advantage for the listoriented question types; and that multiple-text, best passage display would have an advantage for the comparison-oriented question types. The two interfaces were compared on effectiveness, usability and preference measures fo...
Nicholas J. Belkin, Amymarie Keller, Diane Kelly,