Design of Knowledge-Acquisition Tools

Knowledge-acquisition tools can be based on several models-for example, models of cognition, models of knowledge representations, and models of problem-solving methods. Musen [14] describes three basic conceptual models for interactive knowledge-acquisition tools: symbol-level, task-oriented, and method-oriented conceptual models. The supportive power and the scope of a knowledge-acquisition tool follow from the model supported. Moreover, we can conceive many models for metatools. The essence of a metatool is the model for the target knowledge-acquisition tools that the metatool supports. We use the term for such specification models for knowledge-acquisition tools. We divide the development process for knowledge-acquisition tools into three major stages:

1. Knowledge-acquisition analysis: In this stage, the developer analyzes the domain and the task, and outlines the requirements for the knowledge-acquisition tool. The developer must acquire domain knowledge manually to determine the features required by the knowledge-acquisition tool.

2. Tool specification: In this stage, the developer models the knowledge-acquisition tool. When a metatool is used, the result of this stage is a specification that conforms to the metaview supported by the metatool; otherwise, the result is a regular software specification.

3. Tool implementation: In this stage, the developer-or the metatool-implements the knowledge-acquisition tool. Metatools transform the tool specification developed in step 2 to an operational program.

Developers can model domains independent of the problem-solving method through specification approaches, such as knowledge-level analysis [16] and ontological analysis [1]. However, in the general case, the transformation from such models into a high-quality knowledge-acquisition tool is nontrivial. Target knowledge-acquisition tools must be designed by developers in cooperation with the tool users (e.g., domain experts). We seek appropriate knowledge-level descriptions for domain-oriented knowledge-acquisition tools (rather than descriptions of domains or descriptions of problem-solving methods), because knowledge-acquisition tools are different from target systems. Our research objective is to develop metaviews for domain-oriented knowledge-acquisition tools that are general; that is, they are not restricted, for instance, to a particular problem-solving method.

We have developed a metaview for target knowledge-acquisition tools that comprises generic building blocks for knowledge-acquisition tools-for instance, generic user-interface components for interactive knowledge editing, generic knowledge representation structures for internal use in the knowledge-acquisition tool, and generic knowledge-base generators that produce target knowledge bases. This abstract-architecture view incorporates architectural components of knowledge-acquisition tools at an abstract level. The abstract-architecture view provides the developer with a conceptual model of the target knowledge-acquisition tool that is based on the tool's architecture. Developers can specify a broad variety of knowledge-acquisition tools by specifying such building blocks and by defining the relationships among them. A metatool can automate stage 3 by transforming such specifications into an implementation of the knowledge-acquisition tool.