The extension and application of Swet's theory of information retrieval
The thesis comprises (1) 8 critical interpretation of Swets's contribution to information retrieval, (2) development (i.e. "extension") of the formalism, as so interpreted, and (3) a description of an experiment that identifies hypotheses consistent with the extended formalism. The early sections of the thesis place the original contribution by Swets in the contexts of both signal-detection theory and information retrieval theory. It is then argued that as the original theoretical contribution is ambiguous in key respects, an interpretation of it is necessary. The interpretation given constitutes an initial development of Swets's work but other developments, not simply a consequence of the interpretation of the original description by Swets, are also put forward. The major one of these is the explicit incorporation in the formalism of logical search expressions. Elementary logical conjuncts of search terms are seen as (1) being weakly ordered by "document ordering expressions", and (2) having probability-pairs attached to disjunctions of them defined by the ordering. A major part of the thesis is the identification of novel hypotheses, expressed within the extension of the original formalism, which relate to triples of: (1) instances of information need in medicine, represented by prespecified partitionings of a medical-literature data base (MEDLARS), (2) an analytical document ordering expression, and (3) an algorithmically-derived set of terms characterising the information need. An enhancement is suggested to data base management programs that at present employ only user-specified logical search expressions by way of search input, this enhancement stemming directly from the extension of the original formalism. The broad conclusion of the thesis is that when the original contribution of Swets is suitably interpreted and extended, a robust, hospitable conceptual framework for describing information retrieval at the macroscopic level is provided.