Generation of model-based knowledge-acquisition tools for clinical-trial advice systems
暂无分享,去创建一个
To construct knowledge-based systems, developers must create models of the problem-solving behavior of experts. This modeling process, termed knowledge acquisition, is laborious, and thus impedes the dissemination of expert-systems technology in all application areas, including medicine. This dissertation demonstrates ways in which computer-based tools can accelerate knowledge acquisition by aiding system builders in the creation and application of expert models.
In many application areas, there are related tasks for which advice systems might be useful, and for which only one general model is needed. One such domain is that of clinical trials--formal medical experiments in which alternative therapies are compared. Although different clinical trials require different treatments, in various medical specialties there are classes of clinical trials to which the same generic model of treatment planning applies. In addition to similarities among the tasks to be performed, the computational methods that an expert system might apply to carry out those tasks are the same.
This dissertation presents a methodology to facilitate knowledge acquisition for advice systems in domains where end users require multiple, but related, knowledge bases. System builders use computer-based tools to create models of the tasks to be performed. Once those task models are designed, other computer-based tools are generated automatically to allow application specialists to enter directly the details of particular tasks. I have demonstrated this methodology in a system called PROTEGE, which helps its users to build models for tasks that can be solved by successive refinement of skeletal plans. In the domain of clinical trials, system builders use PROTEGE to create models of particular classes of treatment plans. PROTEGE then produces custom-tailored, graphical tools with which expert physicians define the details of particular clinical trials within each class. The physicians' specifications are automatically translated into the knowledge base of an expert system.
The methodology separates the problem of modeling an application area from that of entering content knowledge. This division of labor expedites the creation of advice systems for application areas in which multiple large knowledge bases are required for analogous domain tasks.