Manhattanville College expert academic advisor---preliminary report (abstract only)

摘要

Manhattanville College is a small, liberal arts college located in Westchester County, New York. Every student plans his or her program each semester with an academic advisor. Unfortunately, this advisor is not always from the department of the student's major. We see advising as a serious academic problem. We decided that it would be useful to have an expert system that could assist in advising. At a small college, student-faculty interaction is very important. The intended use of the system is not to eliminate this contact, but to enhance it. We view the system as an expert aid to advisors and students, not as a way of replacing human advisors. Most of the development of expert systems that give advice appears to us to be in the areas of business, manufacturing, and medicine. The extensive project Athena program at MIT, for instance, covered over sixty academic computing projects, but did not mention computer assisted support for academic advising.[1] This is an area for fruitful study, we believe.

We will report on the first results of an on-going project to develop an expert system that gives advice like that given by an "expert human academic advisor" on courses to be taken to satisfy a student's major requirements. The system could be used by students planning their schedules for future semesters and by advisors outside the student's major department. We decided that the system should run on IBM PC's and that systems developed should run as stand-alone, turn-key systems. After a survey of the literature on expert system development we chose to use an expert systems shell to develop our package. We felt that the Arity Expert System Development Package c and Arity Prolog Compiler c provided a powerful shell to do the initial development as well as the flexibility to enhance the system by including code written directly in Prolog.

The system that we currently have running gives advice to Mathematics and Computer Science majors. We will describe the knowledge acquisition and system building interaction that is required for the development of this type of academic application. In particular, we will discuss: the modular development approach; development of course recommendation trees; development of taxonomy files; translation of course recommendation trees into a rule base; programming the expert system to work with these rules; general rules and mop-up rules; the relationship of the order of the rules in the rule base to the order of the questions asked by the system; testing and debugging the system; the problems of integrating this approach into the traditional system of academic advising.

The creation of this system was unusual in that we played two roles in its construction. We were both the knowledge engineers and the experts. We often found ourselves alternating between these two roles. We discovered that frequently one or the other of the roles was difficult to understand. We expect that our experience will be directly transferable to knowledge engineering in areas other than areas of our own expertise. Our plans for future expansion of this system include the programming of advice for other departmental majors. While many of the techniques and development approaches that we devised appear to be directly applicable; we have not yet developed clear heuristics as to how to ask the right questions of outside experts. A major expansion of the system will allow us to focus on the area of knowledge engineering.

We have already seen that the size of the expert system we have developed for Mathematics and Computer Science is quite large. An immediate question is whether a college wide expert advising system would best be implemented as a single expert system or as a collection of expert system modules for various areas of the curriculum. If the latter approach is taken, we must determine the optimum module size so that performance during individual consultations does not degrade unacceptably.