INTELLIGENT DOCUMENT INFORMATION RETRIEVAL SYSTEM FOR DETERMINANTS OF MEDICATION RESEARCH LITERATURE COLLECTION.
Dr. Steven B. Schoenly, Mr. Jan G. Wilms, and Mr. Mustapha Sarji,
Department of Computer and Information Science, University of Mississippi;
Dr. Mickey C. Smith,
Department of Health Care Administration, School of Pharmacy, University of Mississippi.
The subject domain "determinants of medication" is an interesting one for information retrieval system development. Literature concerning the selling, advertising, marketing, prescribing, purchasing, and taking of medications forms a well-defined and rather small subject domain, but is scattered over a broad range of publication sources with widely varying publication practices. Furthermore, this subject domain is not well covered by any single existing medical database. An established collection of documents concerning determinants of medication forms the basis for an intelligent document retrieval system, the prototype implementation of which is underway at the University of Mississippi Department of Computer and Information Science, working ln cooperation with the Department of Health Care Administration in the UM & school of Pharmacy.
The prototype system represents an attempt to implement an operational information retrieval system that provides enhanced performance by utilizing artificial intelligence techniques in connection with conventional information retrieval techniques. One approach to the adaptation of artificial intelligence concepts and techniques to the area of information retrieval system design is to attempt to model the intelligent information retrieval system after the expertise demonstrated by the trained human information specialist,who may have little prior training in the subject area covered by the document collection in question. This approach allows the design of the system to be built on a foundation that utilizes well-known information retrieval techniques, while holding open the possibility that retrieval performance may be enhanced by indexing, processing, and user interface mechanisms that are not found in conventional information retrieval systems. A natural language user interface can eliminate the need for a formal query-formatting syntax, and can effectively eliminate the need for the user to express queries as logical document set manipulations of specific terms, without requiring infallible detection of syntactic errors or semantic ambiguities ln the user's query statement input. Retrieval processing mechanisms, based on conventional inverted index files that provide access keys for all aspects of the information content of the documents included in the collection, can manipulate multiple term weighting information sources (coming from the user interface component, as well as being embedded in the index files themselves) to retrieve documents, in response to queries,that match the queries inexactly or approximately. Communication between the user interface component and the processing component of the system is based on representations of queries as idealized documents, or patterns of information content, similar to a greater or lesser extent to representations of documents actually stored in the collection. Traces of queries submitted by the population of users allows the system to adapt its performance roughly to the anticipated retrieval expectations of individual users, or classes or users.
The prototype system is being developed as a research tool that can be readily adapted to other subject domains, because processing mechanisms and "knowledge sources" are not tied specifically to the subject domain of determinants of medication. However, the system is intended for prompt, practical implementation ln an operational mode for routine use by specialists in the research covered by the subject domain used for the prototype design work.