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. 296 An experiment: search strategy variations in SDI profiles (3) T4 Assign weights to terms In assigning weights subjectively to individual terms, values in the rangc 1-10 were used but where a single concept seemed paramount to the query the scale was extended to cover weights of 1-20. The facility for utilizing a wider range of weights was available (-999 to + 999) and in an operational system much higher positive and negative weights could be used for ranking highly or excluding altogether items in certain journals or languages, etc. An interesting characteristic of assigning weights subjectively to individual terms is that almost invariably one finds oneself separating the terms into groups (task T2), mentally ordering the groups into their relative importance (task T5), and setting imaginary threshold weights in a manner very similar to constructing a logical search. It is not suggested that one necessarily has to formally perform tasks T2 and T5 in order to assign individual weights to a single list of terms; only that it is difficult to proceed directly from task Ti to task T4 without thinking in terms of tasks T2 and T5, and even simulating T3 (assigning boolean equations). A reservation concerning subjective weighting of terms is its possible lack of appeal to individual users interested in compiling their own profiles. No doubt some individuals would delight in the facility; others might be quickly frustrated by the problems of unanticipated homonyms, etc. (4) T5[OCRerr]Arrange groups in order of importance Normally it is a fairly quick task to arrange subjectively the groups of terms in order of their importance to the query. When `powers of 2' weighting is being used this task becomes quite crucial to strategy GTWC. The feeling persisted that `powers of 2' weighting would be better suited to controlled- rather than free-language searching, i.e. it would operate better with greater pre-coordination of terms. If this is so then strategy GTWC may have suffered somewhat from the approach advocated for task T2 which was to divide into as many concepts as possible. (5) T6[OCRerr]Assign weights to groups The same general procedure was followed in assigning weights subjec- tively to the groups of terms as was described above in (3) for the individual terms. Relevance assessments An important factor when considering the mechanics of the experiment was the role of the user group. It seemed desirable that they should operate in a `near-real' situation but at the same time it was necessary to have as many documents assessed as possible to ensure the validity of the recall figures. The users were not, and did not need to be, aware that a number of profile versions (representing the different search strategies) had been compiled from each statement of interests. On a particular search run the separate outputs from all the profile versions prepared from the user's original statement were merged to produce a single set of notifications without duplicates in random (document number) order. This was not only the most convenient procedure for the user but was also methodologically necessary in