Expert Retrieval Assistance Development and Experimentation
暂无分享,去创建一个
The development of enhancements to the CONIT retrieval assistant system leading toward an expert system version are described. Enhancements included a combined menu and command interface allowing users to get assistance in selecting appropriate search modification tactics and have recall estimates calculated by CONIT. Experiments on a partially completed version of this system are summarized. Our conclusion from the development, experimentation, and analyses is that expert retrieval assistance via techniques we have proposed does provide a potential for vastly improved search effectiveness for a wide class of users. However, the development effort to realize this potential is large and, as yet, only partly completed. The particular efforts most needed to realize the potential are included in our design for the stage 2 Expert CONIT system. 1. Background We and others have pioneered in research ([MARC83b], [WILL86]) to assist retrieval system users, especially inexperienced end users, by way of the mechanism known as an intermediary (or front end or gateway ) system (our experimental intermediary system was named CONIT). Despite the apparent success of this research, including the operational systems developed by the new field created therefrom ([LISA84], [HUSH86], [WEIN86]), we realized that much more powerful retrieval assistance effectiveness was possible. Preliminary explorations along these lines were carried out (see e.g., [MARC81] and [MARC83a]) and led to our investigation of what we have termed an "expert retrieval assistant." Our recent work in this investigation, described in this paper, has focussed on three main areas: (1) detailed design and partial implementation of an expert version of our CONIT retrieval assistant; and (2) preliminary testing and analysis of CONIT in its "standard" and "expert" modes; and (3) partial conversion of CONIT to a workstation environment. 2. Expert CONIT Development The expert retrieval assistant, while seeking a new level of intelligence and sophistication, depends for its success on a number of kinds of functionality that we have already developed in standard CONIT including (1) a common command language; (2) a knowledge base for the distributed, heterogeneous retrieval systems built from production rules; and (3) a search methodology based on automatic generation and execution of search strategy derived from a user's natural language topic expressions. Expert CONIT design has, then, been conceived as a series of encapsulating elaborations of standard CONIT considered as a basic core. The first elaboration consisted of adding a "construction" or "meta" level in which individual searches and whole search strategies could be constructed prior to their actual execution in one or more files. The construction level was also coordinated with the ability to save search strategies from session to session in search catalogs for individual or shared community use over time.
[1] Martha E. Williams,et al. Transparent information systems through gateways, front ends, intermediaries, and interfaces , 1986, J. Am. Soc. Inf. Sci..
[2] Richard S. Marcus,et al. An experimental comparison of the effectiveness of computers and humans as search intermediaries , 1983, J. Am. Soc. Inf. Sci..
[3] Man-Kam Yip,et al. An expert system for document retrieval , 1981 .
[4] Martha E. Williams,et al. Transparent information systems through gateways, front ends, intermediaries, and interfaces , 1986, J. Am. Soc. Inf. Sci..