ABSTRACT
An open-ended classification system for online databases was developed with nine attributes, each one of which has a controlled vocabulary schedule of categories. The first eight attributes involve miscellaneous coverage features and the ninth is a new subject classification using a three-level viewpoint schedule. Its categories were obtained by pooling and simplifying the subject indexes of three prominent database directories (Cuadra, KIPD, Williams), a gateway service (EasyNet), an online vendor’s term frequency index (DIALINDEX), and an online strategy textbook (Hoover). This method of using published accounts captured the varying concepts deemed important by those already familiar with online database selection. Every category of each attribute was assigned a rank, which is a figure of its relative merit, albeit subjective, in terms of its generality. The classification system was used in creating a relational database of online databases that is the factual basis for retrieval.
A microcomputer program was written using the GURU (trademark) artificial intelligence development system. It consists of the internal database of all the attributes and ranks for the online databases considered and a script performance file that calls several modules. Each module has conventional context-sensitive help messages and rule-based expert system advice. The “user modeler” internal expert helps the users delimit their time, language, and geographic 208coverage requirements and specify the extent and depth of coverage needed. The “question clarifier” internal expert helps users discover the underlying viewpoint of their need. The “searcher” internal expert gets four sets: one that the user asked for, and three more with successively broader viewpoints, including principal databases in the user’s field of expertise. The “evaluator” internal expert analyzes the postings and guides the user, if requested, to the appropriate set to inspect. The inspection consists of reviewing uncontrolled textual descriptions that provide unique or special features of each database. The “ranker” internal expert provides an ordered list of those databases judged pertinent by the user.
Alternate methods of access to this expert system’s knowledge are provided by the “browser” internal expert, which allows free-text search of text files and indexes. Users may contribute comments with the program’s word processor and they may add their own data. Each session can be saved for continuing or restarting at a later time.
The authors conducted this work at the Department of Library Science and Information Science, Åbo Academy, 20500 Turku, Finland, under a grant from the Academy of Finland. Current address of Rodes Trautman is Library Scientists, 7266 E. Camino Valle Verde, Tucson, AZ 85715; that of Sara von Flittner is Åbo Academy, 20500 Turku, Finland.
The authors are appreciative of the encouragement given by Mariam Ginman and the expert editing assistance of Phyllis Chen.
