|
|
© 1997 Copyright on this material is held by the authors.
The system design process can receive input from the user population before and after system deployment. For example, participatory design [3] includes the work practices of the targeted user population as a part of the design process. Alternatively, other systems respond to the practice of users at run-time, i.e. during deployment, either by adapting automatically to the habits of the user (adaptive systems) or directly by the user (adaptable systems) [2].
Our method of system development is to gather input for system re-design both during and after deployment. We want to take advantage of the notion that a computer tool can record a database of its usage which can then be used as input to designers in the re-design phase. This data is consulted during re-design to provide guidance for system modification. We refer to this technique as Participatory Adaptation. During run-time, we collect the domain and user-interface usage data of experts. It is often hard to classify usage data as either domain- or user-interface-specific, for example when a domain event triggers a user-interface action. Allowing the two types of data to intermix has the potential to be a powerful technique of system adaptation, cutting across both the sign and tool functions of the system. **
Typical HCI systems address the sign function -- the user interface -- while traditional AI systems examine the tool function -- the user-domain relationship. One of the underlying propositions of our work is to build systems that jointly adapt both the sign and tool function of a system.
The user interface for our system, called Interactive Mover's World (imw), is shown in Figure 1.
Figure 1
The graphics area illustrates the problem state. The row of buttons above the graphics area activate a series of pop-up windows that the user can manage in order to access various features of the interface. Experts exhibit patterns of usage through popping up and moving around windows, triggered by domain and interface events.
The manipulation of the interface is purposefully complex. Users are intentionally bombarded with sources of information and choices to make. The idea is to simulate an environment that is rich in data, both stagnant and changing in real-time, and to require cooperation between users in order to solve a problem.
The user is required to manage the pop-up windows in order to plan moves, send plans to an executive process that synchronizes actions among users, communicate with other users, examine a database of object properties and access a help facility. All user communications occur via imw's communications facility, which provides a set of canned messages pertinent to the various situations a user might encounter.
A second user interface is being developed for comparison, as shown in Figure 2. The main difference between the two interfaces is evident in the graphics area: the first interface is more intuitive, using squares and stick figures to represent objects and movers. The second interface is entirely symbolic and requires the user to learn a mapping between colored symbols, objects and movers.
Figure 2
As a first step in fitting the usage data, we are currently implementing several techniques: the COBWEB incremental classifier [1] is being used as a basis for comparison; the MULTICASE [5] is a technique that may prove effective with our data. The MULTICASE is a method for merging multiple episodes of routine activity into a single representation. Our plan is to use the MULTICASE to represent the base of expert activity, which can then be used to recognize subsequent user behaviors, for both sign and tool functions.
Figure 3
Our long-term goal is to be able to recognize and exploit the methods of experts by capturing and analyzing usage data generated during the run-time of a complicated problem solving system. Our plan is to use this run-time data as a basis for adapting the system so as to improve the performance of novice users. We plan to test this hypothesis experimentally by comparing the efforts of two sets of novice users: those who work with the adapted system and those who work with the original (unadapted) one. Our belief is that the re-designed system, adapted through the participation of expert users, will serve to enhance novice performance.
2. Oppermann, R., ed. 1994. Adaptive User Support. Lawrence Erlbaum Assoc.
3. Schuler, D. and A. Namioka, ed. 1993. Participatory Design: Principles and Practices. Lawrence Erlbaum Assoc..
4. Vygotsky, L. S. 1978. Mind in Society. Harvard Univ. Press.
5. Zito-Wolf, R., and Alterman, R. 1992. Multicases: A case-based representation for procedural knowledge. In Proceedings of the Fourteenth Annual Conference of the Cognitive Science Society.
** It was Vygotsky [4] who noted that any mediating artifact has two functions: sign and tool. In the case of computer systems, its sign function provides the basis of communication with the user, and the tool function, a set of methods for manipulating the data.
|
|