IRE
Information Retrieval Experiment
Actual tests
part
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.
Part 3
Actual tests
In this section individual tests are described and evaluated. Where Part 1
discussed general problems of testing and Part 2 those of types of test, the
intention here is to focus on individual experiments and what can be learnt
from them. The chapters taken together show what the substantive concerns
and methodological approaches of retrieval system tests in the last twenty
years have been, and so throw light on the state of the art from a third point
of view. In particular the tests described indicate how, and how far, testing
experience has grown in the field as a whole.
The four chapters illustrate the themes of the previous sections in different
ways. In the first chapter by Sparck Jones the whole development of retrieval
system testing, and specifically experimentation, over the last two decades is
surveyed, with references on the one hand to major tests and on the other to
average or typical ones. Sparck Jones' second chapter is devoted to the
Cranfield Project tests, particularly Cranfield 1 and 2, which are widely held
to have been exceptionally important. These key tests are presented chiefly
in their own terms, and looked at through the eyes of contemporaries. The
chapter thus brings out the contribution of Cranfield 1 and 2 to retrieval
experimentation at the time that they were conducted; but their longer term
contribution to retrieval system testing is also considered. In the third chapter
Evans gives a full-dress account of a single experiment studying the
performance effects of different search strategies. This is a detailed case
study showing how a single test was originally approached, and how it
appears retrospectively. Finally, the largest and longest retrieval system
testing project, the Smart Project, is reviewed by Salton. As system testing is
expensive and individual tests are too often incomparable, there are great
advantages in facilities for long-term testing. Salton's discussion of the way
these facilities were provided and used by the Smart Project, and also of the
disadvantages which may accompany the advantages, provides a conclusion
for the book which reiterates the point made in the Introduction: namely that
retrieval system testing is difficult.
211