Maximizing the utility of existing automated information retrieval systems is still an outstanding problem of information science. Systems performance depends on three major factors: (a) the proper interpretation of queries posed to the system, (b) the information representation of the documents in the information store, and (c) the strategy by which the documents relevant to a query are identified. Clearly, the system's ability to establish the relevance of documents to a given query is crucial for the system's performance and underlies the efficacy of each retrieval transaction. Unfortunately, relevance cannot be formally defined nor precisely measured. Yet it is the key issue of the retrieval problem. Operationally, two different approaches have been used independently of each other, namely, retrieval by descriptors (semantic relevance), and by citations (pragmatic relevance). However, the relationship between these two approaches is not known, albeit there is evidence that the outputs for the same query based on each type of relevance are dramatically different. Therefore, a significant research gap exists suggesting the study of the combined use of these two modes of relevance, as they are tested with a number of search strategies and query analysis methods. A set of algorithms incorporating various combinations of question analysis procedures, documents representations, and search strategies can be developed and implemented as an add-on device on existing systems such as MEDLINE. Such retrieval enhancement can preserve, promote, and maximize our capital investment in the creation and organization of biomedical databases.
