An investigation of the optimization of search logic for the MEDLINE database

Abstract

One of the key variables in the optimization of the retrieval process is the logic imposed on a set of query terms. If the query is small, a combinatorial algorithm can be employed to identify search expressions having an optimal logical form. An experiment is described in which this was done for queries expressed against MEDLINE, for a variety of criterion variables. The method employed is useful not only for assisting in identifying optimal logical forms, as demonstrated, but also as an experimental control device to assist investigations into the effects of varying the set, and number, of search terms. The experiment also suggests several novel properties of effective searching against MEDLINE, for example that searching with four MeSH terms is likely to be more successful than searching with a lesser number of terms, provided search logic is optimal. The latter result is, surprisingly, true for both Precision and Recall, i.e., a tradeoff between the maximally attainable values of these variables fails to hold when the number of search terms varies in this range.

An investigation of the relative influences of database informativeness, query size and query term specificity on the effectiveness of Medline searching

“…However, more specifically, since the operationalisations of relevance judgements required it, the database was chosen to be a specific section of Without such time-scoping or 'sectionalising' of the database, Precision will be assessed in the experiment as being less than it should be, assuming the experiment is simulating users' relevance judgements. (2) The method used to operationalise relevance judgements. Here, the notion of 'a set of relevant documents' was operationalised as the set of documents cited by the author(s) of a medical review paper.…”

“…In adopting such a procedure, it is necessary to make careful, additional decisions as to just what 'optimising by Recall' (say) actually means. The considerable problems that exist here, and the supplementary rules used, are detailed in [2]. For example, it is meaningless (or rather, trivial) in practical terms to identify search expressions that maximise Recall at the price of retrieving the entire database, by 'ORincluding' the all-negated elementary logical conjunct of the query terms.…”

“…Earlier related work, which included a treatment of the mappings of all possible search expressions to the R x P outcome space, was reported in a review paper by the author [1], and a fuller analysis of the practical problems of assessing such variation from a logic perspective has been offered by Heine and Tague [2].…”