111-7

Figure 2 also gives the postings of the nine Mesh terms used in the search

formulation plus two more terms from the request not used in the search for-

mulation.  Assuming that the search formulation included all the terms likely

to be connected with the requestort 5 need, it can be seen that, ignoring

the logical combinations asked for in the search formulation, 8,ooo to 10,000

documents in the file are associated with the search term chosen, or, `i~re

simply still, regarding `tCerbrospinal fluid11 as an essential notion, 1162

associated items are in the file.  The user' 5 second need would almost

certainly be satisfied by some of these 1162 items, but it might have been

possible to have satisfied both the user's first and second needs in a single

search in the following manner:  with `tcerbrospinal f1~id1' identified as a

key notion, all the ll62 documents posted with that term would be retrieved,

and would then be presented to the user in some `1ranked'1 order of probable

relevance.  A satisfactory result for this process would be obtained if

the U references found in the Medlars search were to appear at the front

of this ranked list (the user identified 10 of these ll as relevant).  ~f such

a system could be provided, the user' 5 second need could have been satisfied

merely by an examination of more than the first 11 documents in the search

output.

         A second example of problems of this type lies at the opposite

extreme:  a request on the effects of drugs and pesticides on the bone marrow

of man and animals resulted in 1,235 retrieved documents.  Since the set

almost certainly contains too many items, a ranked output would again

provide a solution, allowing the requestor to examine only as many documents

as he desired.   Searches in Medlars are frequently made in three matching

strengths in an attempt to meet this latter problem.