Browsing by Author "Richard S. Marcus"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    Design questions in the development of expert systems for retrieval assistance
    (2007) Richard S. Marcus
    A number of efforts have been ongoing to investigate the prospects for expert computer systems that would match or excel human experts in providing assistance to users of retrieval systems. In this paper we contrast such efforts in terms of the models for retrieval and assistance they subsume and in terms of the techniques for performing effective retrieval and for developing expert assistance systems. Further, we state and attempt to support three premises: (1) to provide a truly comprehensive expert retrieval assistant requires a very extensive knowledge-base development; (2) there are significant questions concerning retrieval models and assistance techniques which need to be resolved in developing such expert systems: and (3) although expert retrieval assistance development is difficult, it shows promise for deepening our understanding of the retrieval process from a basic scientific viewpoint as well as for improving search techniques themselves. In support of these premises we discuss some of our recent experiences in the development of our CONIT experimental retrieval assistance system.
  • Thumbnail Image
    Item
    Development and Testing of Expert Systems for Retrieval Assistance
    (1985) Richard S. Marcus
    Investigations have continued in the development and testing of experimental expert systems, under the generic name CONIT. for assisting all classes of users 10 access and operate mutliple, heterogeneous online retrieval systems. New experiments have now been run in a more "open" and "realistic" environment — users operating an enhanced CONIT at times and places of their convenience (e.g.. their own offices) but with definite monetary constraints. These experiments indicated that, while previous results showing easy learning with some relevant retrieval still obtained, the time and cost pressures did have a negative impact on the effectiveness of the searching. Some of the improvements in the enhanced CONIT showed promise for increasing effectiveness and efficiency in lime/cost pressured environments. Major additional improvements were deemed possible and highly desirable in several areas including (1) a true integration of menu and command modes, and (2) a more directed, dynamic, and comprehensive capability for planning assistance and search evaluation — both before and after searches and contemplated search modifications were run. We have designed and partially implemented a modified CONIT intermediary system to evaluate these additional improvements. The new integrated command/menu interface leaches commands by dynamic example in response to menu selections while allowing users to give command or menu inputs at their own discretion without explicitly changing models.