An investigation into the application of expert systems to reference work

Abstract

Expert systems have been used in Library and Information Studies (LIS) for classification, cataloguing, indexing, book selection, online searching of databases, and in reference work. This thesis reports an investigation into the application of expert systems in reference work. The thesis reviews models of the reference process. The reference process can be viewed as having four stages: query clarification, source selection, searching, and answer submission. Jahoda suggests a model of source selection in which a reference question is classified by identifying the wanted, the information that the enquirer is trying to find out, and the given, the information that the enquirer already has. The wanted and given can be used to match queries with appropriate reference sources. The thesis also reviews implementations of Reference Advisory Systems (RASs), expert systems used in reference work. RASs concentrate on the source selection stage of the reference process. Source selection is less reliant on broad general knowledge, and on knowledge of natural language and human relations. Existing implementations of RASs are largely limited to menu-based interfaces. While most RAS implementations have been rule-based, there are indications that frames may be an appropriate knowledge representation method for RASs. Two implementations of RASs for New Zealand reference work were developed as part of the investigation, using the EXSYS expert system shell. Kiwinet Advisor is a small system that selects appropriate databases from the Kiwinet online information service. A larger system, the New Zealand Reference Advisor (NZRef) uses the Jahoda wanted/given model to select between 118 New Zealand reference sources. An evaluation of NZRef with practising reference librarians indicated that it was a useful aid to a library assistant answering reference questions about New Zealand.