Toward an integral multi-file on-line bibliographic database

Abstract

The work described represented an attempt to lay the basis for an integral multi-file bibliographic database and, in particular, to derive a common retrieval vocabulary for data bases received from many different sources in a single carrier language. The creation of an experimental database of some 44 000 references with abstracts, as originally announced in NASA STAR during 1973 and 1974, is described. This data base could be searched either by the classical controlled voca bulary technique or by means of an improved natural language technique using terms in the titles and/or abstracts, or any combination of these. This improved technique was based on three discrete aspects: the development of an index file of uniterms from the natural language by means of an iterative, machine-aided procedure; the incorporation of meaningful phrases culled from the same source; and the derivation of an RT-analogue, on the basis of statistical term association, called Associated Concepts. The experimental database was submitted to a comparative evaluation of controlled and natural language retrieval performance by an external group directed by Prof. Cyril Cleverdon and com prising expert documentalists from several large European information centres with long experience of on-line search ing ; the results of this evaluation are reported.