CRANV2
Aslib Cranfield Research Project: Factors Determining the Performance of Indexing Systems: Volume 2
Main test results
chapter
Cyril Cleverdon
Michael Keen
Cranfield
An investigation supported by a grant to Aslib by the National Science Foundation.
Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government.
- 159 -
Section 8 Controlled terms
Recall devices
The investigations on the controlled term languages were carried
out on two sets of questions and two collection sizes: the 42 questions
(subset 2) searched on the 200 document collection (subset i),
and the 77 questions (subset 7) searched on the 350 document collection
(subset 2). Six recall languages were tested, and the effect of each
recall device is seen in the tables and plots given in Figs, 4. 800TP
4.805TP, tested with a search rule A. A plot of the six curves
is given in Figure 4.806P. The recall devices are shown in Fig. 2.7 (p, ll)
Search Rules
Further comparison ef the six index languages is given with the
tests made using Search Rule E on both the 42 questions with the
200 document collection and the 77 questions with the 350 document
collection. The results are presented in Figs. 4.810TP - 4,815TP
and 4,820T - 4,825T, with plots covering the various index languages
as Figs. 4.816P and 4.826P. With Search Rule E all the combinations
of acceptable terms were selected for each coordination level. Examples
of search formulations for a number of questions are given in Appendix
8.1.
Tests with Search Rule 17' were done on index languages III. 5.a and
II1.6.a with 42 questions on the 200 document collection. This search
was superimposed on Search E, and the results are presented at the
various coordination levels according to the number of basic terms as
apart from related terms. Thus, at a coordination level of 4, the
tables (Figs. 4.850T and 4.851T) show the results when all terms
were basic terms, where one term was a related term, where two terms
were related terms and so on. It is obvious that when all the terms
are basic terms, then the results must be the same as for index language
III. l.a With Search E; when all the terms are related terms, then the
results must be the same as for the corresponding index language with
the basic Search E. Therefore in the plot 4.850P, the two main
curves represent index languages III.l.a and III. 5.a with Search E and in
Figure 4. 851P the main curves represent index languages III.l.a and
III. 6.a, again with Search E. Tile additional results obtained with Search
F now produce a series of secondary curves at each coordination level
which span the main curves.
Precision Device
The precision device of weighting was tested, in which search questions
were weighted and tested on the most exhaustive index language, using
the weights assigned in indexing. Figs. 4.830TP and 4.831TP give
results for languages III.l.a and IIi.6,a respectively, and in the plots
a comparison is made of the weighted and unweighted searches, carried
out with search rule E.