CRANV2 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. - 131 Section 5 Search Rules The first search rule investigated is type B, which, as explained in Chapter 2, was used to provide a minimal amount of intellect prior to making the test of precision devices. The sets of questions (subsets 4 and 5) and languages (I.l.a and 1.6,a) that were used to test precision devices are investigated. Figs. 4. 500T - 4.503T give tables of results for search A, this being the basic type of search used in presenting the results on, for instance, the recall devices. This particular series of results has to be given because of the different set of questions which was used for testing Search B; the results of these latter searches have already been presented in Figs. 4.300T to 4.303T in Section 3. Graphs comparing the curves of search A with search B are given in Figs. 4.504P to 4.507P. Search Rule C demanded a selection of the original starting terms taken from the question. This eliminated such terms as Problem, Influence, Comparison, etc., and the result was to reduce the number of starting terms in every question, Whereas originally, in this subset, each question had seven starting terms, when search rule C was in force, the number varied from two to six. Search Rule D retained the terms selected for search rule C but imposed restrictions concerning the combinations of terms that would be accepted at every coordination level, thereby eliminating non-sensical combinations. For search rules C and D a set of 20 questions (subset 8), all taken from the 35 questions having seven starting terms (subset 1) is used, searched on the 1400 collection. Languages I.l.a and 1.6.a are tested and Figs. 4.510T and 4.511T give the results with search rule A. Results for searches C and D were totalled by method 1B, (see chapter 3, page 62), since the ordinary strict coordination level method gave unsatisfactory results due to the small number of questions and variation in sta[OCRerr]ing terms that resulted from the rules used in the searches. Figs. 4.512T and 4.513T give the results of search C, and Figs. 4.514T and 4.515T the resu!ts of search D; because of the different totalling methods the tables of results have to include the additional data which shows the reduce[OCRerr]t number of documents regarded as relevant at the higher coordination levels, and also the resultant change in generality number. Two plots Figs. 4.5161' and 4.517P present the result for searches A, C and D with the two index languages. Although the comparison between search A on the one hand and searches C and D on the other hand might seem to be influenced by the different t0talling method used, all possible methods were tried and that used was found to be the most satisfactory.