IRE Laboratory tests of manual systems chapter E. Michael Keen Butterworth & Company Karen Sparck Jones All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, including photocopying and recording, without the written permission of the copyright holder, application for which should be addressed to the Publishers. Such written permission must also be obtained before any part of this publication is stored in a retrieval system of any nature. Controlling searching in experiments 147 100 (o) Uniterm 80 Alphabetical U.D.C. a, Facet 0 6C L U 4C 2C 2 3 4 5 >6 Subsearches bc (b) 80 6C 4C 2C Uniterm Alphabetical ___________________ Facet UDO 0 C, C, U 1 2 3 4 5 >6 Subsearches Figure 8.5. Cranfield I results from two search rounds, calculated from Tables 3.6, 4.8, and 4.9 in Cleverdon2. (a) Search round two, (b) search round three prompted an additional subsearch then this was allowed and used to create a modified version of all earlier searches. The results of round three were taken to be very similar to round two, and in fact the similarity is most striking when the new presentation of Figures 8.5(a) and 8.5(b) is compared, showing how recall improved with each new subsearch. The relative positions of Uniterm, Alphabetical and UDC are unchanged by search type: only Facet on round three performs a little better in not dropping so sharply at the third subsearch[OCRerr]not a surprising difference as the free-mode searches of round two offen encountered too many chain index entries for pleasure and round three's fixed strategy rules would force such a search inexorably on to the end. But these plots strongly suggest that