IRE An experiment: search strategy variations in SDI profiles chapter Lynn Evans 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. 306 An experiment; search strategy variations in SDI profiles Our boolean statements were meant to be optimum ones but to have optimuili boolean profiles producing ranked outputs suggests a conceptual contradi[OCRerr] tion. To have used less-than-optimum boolean statements would be no reil test of boolean logic. Costs In a computerized search system the costs of the system may be allocated t(I a number of factors-original database production or purchase; compute operations such as searching, output preparation and printing, profile maintenance, etc.; and non-computer operations such as the time ol information scientists, keyboarding, general clerical, etc. Two of these factors were considered in this investigation, viz. information scientist time and computer search costs. Now, nearly five years later, the values used foi salaries and computer processing costs are dated. Rather than simply update them this section considers in more general terms what was learned in the experiment about the various cost factors and how they related to condition[OCRerr] in an operational system. Intellectual (information scientist) effort Although an accurate record of events in the experiment, the times obtaine[OCRerr]l for profile compilation and modification (pp. 293 and 294 above) arc somewhat artificial in the context of a real system in that: (1) only one analysis/modification stage was undertaken; (2) the times for trivial (but necessary) profile modifications were not recorded; and (3) only the actu[OCRerr]tl compilation/modification times were included, i.e. there is no allowance for `dead-time' (tea breaks, etc.) which in real life might involve an additional 20 per cent or so. To translate the figures on pp.293 and 294 to a more real situation we could: (1) assume an average of 3 significant profile modifications per year; and (2) add 20 per cent dead-time to the above compilation/modification times. If that is done comparative estimates are obtained for the information scientist effort required per query per annum for the different search strategies (Table 14.7). TABLE 14.7. Estimated annual information scientist effort per query Search strategy' Information scientist time (mm) CT 90 CTW 107 CG 107 TWC 110 CGW 125 GTWC 128 GWC 136 B 151 BW 168 Assuming a 35-hour week and 47 weeks worked per year, the number of profiles handled per year by an information scientist working full time would