CRANV1P1 Indexing Procedures chapter Cyril Cleverdon Jack Mills 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. - 40 - CHAPTER 4 Indexing Procedures The function of indexing in libraries and information retrieval systems is to indicate the whereabouts or absence of items relevant to a request. It is essentially a time-saving mechanism. Theoretically, we can always find the relevant items by an exhaustive search through the whole collection (assuming that we can recognize what is relevant when we see it). Since this is economically impossible, the size of the store to be examined is reduced by ciassification, using this term in its very broadest sense, i.e., as the recognition of useful similarities between documents and the establishment of useful document groups based on these similarities. So documents, or document surrogates, are assigned to a limited number of classes according to certain criteria, in particular, their subject content (although in machine indexing, utilizing complete text scanning, this 'limited number' can become very large - as large as the number of significant words used in the text). Search for relevant items is made via these classes (which are classes of documents); only those with a probability of containing relevant items are examined, and the rest (hopefuLly the vast majority) are ignored. Clearly, we need to know as much as possible of the nature of the classes to be recognised, and the degree to which they allow reliable predictions to be made as to the probability of relevant items being included in them. Most library indexes, other than those to imaginative works (novels, music scores, etc.) are aimed ultimately at the retrieval of subject information. Even the great Author-Title catalogues, on which so much care has been lavished, serve for the most part the function of a diagnostic classification, i.e., an author's works are sought in the first piace because they are about a certain subject and his name is a clue to locating it. The popularity of the author-title catalogue rests partly on its precision in retrieval. Classes determined by authorship or title are mutually exclusive; there is almost no overlapping, no ambiguity about them and requests can be met with 100% recall and precision. But they are useless if the author or title is not known and it is this situation with which IN is mainly concerned. So the classes investigated by this project are those designed for searching by subject prescription only. There is one exception to this. Bibliographic coupling (including Citation in- dexing) establishes classes for much the same reason as author-title catalogues, as an oblique way of getting at subject content. Papers which have cited item x are assumed to have some connection with the subject of x. This particular device is dealt with separately in Chapter 7. The terms which are used to express a request or a search prescription rarely coincide exactly with the terms used to describe a particular relevant document; this is likely to happen only at a relatively broad level, when a request may be ans- wered by a treatise or monograph on the subject. For example, in the test Q. 93 read 'What investigations have been made on the flow field about a body moving through a ratified, partially ionized gas in the presence of a magnetic field. [OCRerr] Two documents relevant to this question were 1296 'Waves through gases at pressures small compared with magnetic pressure, 1446 'Waves of a satellite traversing the atmosphere' In both cases the match is very imperfect. It is made only by recognising that the