TL;DR
This paper introduces DOTS, a metatool that automates the creation of specialized knowledge-acquisition tools. By using an "abstract-architecture view," DOTS allows developers to build custom tools for specific domains and experts with significantly less effort than manual development, addressing the high cost and inflexibility of existing tools.
What Problem Does This Paper Address?
Developing knowledge-based systems (KBS) often relies on knowledge-acquisition (KA) tools to extract expertise from domain experts. However, existing KA tools are frequently too general or too specific, failing to adequately support the unique needs of a particular application domain or the individual experts involved. Manually tailoring these tools is a time-consuming and expensive process, often beyond the budget of a single project. This paper tackles the challenge of efficiently generating domain-specific KA tools that are adaptable to individual experts and their conceptual models of the domain.
Methods
The core contribution of this paper is the abstract-architecture view and its implementation in the DOTS metatool.
- Abstract-Architecture View: This is a metaview for specifying target KA tools. Instead of focusing on specific domains or problem-solving methods, it decomposes KA tools into generic building blocks:
- Knowledge Editors: User interfaces for experts to input and edit domain knowledge (e.g., forms, graph editors).
- Knowledge Modules: Internal representations that store the knowledge entered by editors.
- Update Rules: Define how information flows between editors and modules, ensuring consistency.
- Transformation Rules: Used by a knowledge-base generator to convert knowledge modules into the target KBS format.
- DOTS Metatool: This tool allows developers to interactively edit an abstract-architecture specification. DOTS then automatically generates an operational KA tool from this specification. It is designed to be domain and method independent, focusing on the architecture of the KA tool itself.
The development process for KA tools is broken down into three stages:
- Knowledge-acquisition analysis: Understanding the domain and expert needs.
- Tool specification: Modeling the KA tool using the metaview (abstract-architecture view in DOTS).
- Tool implementation: Generating the operational KA tool (automated by DOTS).
Datasets
Not discussed in the paper. The paper focuses on the methodology and the DOTS system itself, using examples from the Sisyphus VT task for illustration.
Key Insights and Ideas
- Shifting Focus from Domain to Tool Architecture: The key insight is that by abstracting the architecture of KA tools, rather than their domain-specific content or problem-solving methods, we can achieve greater generality and reusability.
- Metatools as Productivity Enablers: DOTS demonstrates that metatools can significantly reduce the effort required to create custom KA tools, making them feasible for a wider range of projects.
- Domain and Method Independence: The abstract-architecture view allows for the generation of KA tools that are not tied to a specific problem-solving method or domain, offering flexibility.
- Graphical Knowledge Editing as a Core Paradigm: The system assumes that target KA tools will be based on graphical knowledge editing, where experts interact with editors to build their conceptual models.
- Separation of Concerns: The approach clearly separates the specification of the KA tool's architecture from the actual implementation of the target KBS, allowing for deferred design decisions.
Key Results
The paper demonstrates that DOTS can generate functional knowledge-acquisition tools comparable to hand-crafted ones. As an example, a basic KA tool for the Sisyphus VT task was generated in approximately 9 hours, given a domain ontology. This generated tool performed a similar KA task to the well-known SALT system. The authors highlight that the generic knowledge-base generator, parameterized by transformation rules, provides flexibility in the output knowledge bases.
Limitations and Future Work
- Restriction to Graphical Editing: DOTS is currently restricted to generating KA tools that support graphical knowledge editing. It cannot easily generate other types of tools, such as repertory-grid or machine-learning based tools.
- Need for Knowledge-Acquisition Analysis: While DOTS automates implementation, developers still need to perform a knowledge-acquisition analysis to define the overall tool design and requirements.
- Future Work: The authors suggest formulating metaviews that bridge approaches like PROTEGE and DOTS, offering multiple perspectives. They are also developing DASH, a metatool to support knowledge-acquisition analysis and design from domain ontologies. They also mention the possibility of using production rules instead of instances for representing fixes, allowing for deferred design decisions about knowledge representation.
Summary updated