Knowledge Acquisition最新文献

Developing automated support for any workplace involves analysing a workplace, designing a problem-solving approach and knowledge base, populating that knowledge base with information required by the problem-solving approach, and introducing the new support into the workplace. Each of these development phases produces different components of the solution for supporting a workplace. Existing knowledge-acquisition tools support only a subset of the development phases, and the solution components they generate are not integrated: it is left to the developer to create and maintain a mapping between the different solution components resulting from the different development phases. A current trend in knowledge acquisition is to move towards coherent knowledge-engineering environments supporting the entire solution-development cycle. This emphasizes the need for tools that assist developers with integrating the different solution components produced by the knowledge-engineering environment into a coherent system. This paper introduces such an integration tool: the Active Glossary. The Active Glossary is part of the Spark, Burn, FireFighter knowledge-engineering environment. It assists a development team with describing workplaces and programming constructs so that their similarities and differences are made explicit. The result is an explicit mapping between the outcome of a workplace analysis and the design of a problem-solving approach. The Active Glossary further assists the development team with exploiting the similarities for the purpose of reusing previously defined workplace descriptions and programming constructs for new situations.

This paper outlines some of the problems with using predefined building blocks to specify knowledge level models of problem solving, in particular in the context of the KADS methodology. The definition of the basic building blocks in KADS, the primitive inferences, or knowledge sources, often seems to be inadequate to aid the knowledge engineer in constructing an abstract model of problem solving. We argue that the informal, verbal way in which the building blocks are defined is the cause of this problem, and propose to formalize them to make their semantics clear and to assess the consequences of various modeling decisions. We discuss choices among different formalizations, and show in detail the formalization of one class of knowledge sources.

To make plan-based expert systems more accessible to end users and to support user tasks effectively, we present a system that makes the functionality of AI-planning techniques seem natural and immediately understandable. In particular, we present a task support system with a graphical interaction language for the acquisition and display of plan knowledge, where the intended users are domain experts and novices and where previous computer literacy is not required. Based on existing theories in cognitive science and on our own experimental research, we propose a cognitive model of the users' view of tasks. The model postulates the domain experts' ability to recall relevant parts of self-performed tasks in the application domain. The validity of the model is demonstrated in a paper-and-pencil experiment.

Employing a cognitive systems engineering approach, we use the cognitive model, a stage process model of knowledge acquisition, and requirements from the plan formalism to specify DACRON, a system for plan acquisition and task support. DACRON supports the acquisition of plan knowledge by providing graphical representations of domain entities from the users' point of view. DACRON checks the consistency of specified units and graphically aids the debugging process. DACRON also allows the animated presentation of the planning process and its results. To evaluate the usability of DACRON and the relevance of the acquired and displayed knowledge in application domains, experimental studies involving 39 users were conducted. The studies show that over 90% of the subjects could easily use DACRON to enter knowledge, and 80% of the entered knowledge was relevant and correct. In the case of knowledge display, subjects were able to use the displayed knowledge effortlessly and apply it to solve 95% of the domain problems presented.

Many efforts in knowledge acquisition are designed from a knowledge engineer's perspective and as a consequence fall short of allowing experts to elaborate successfully their own situated knowledge. Knowledge engineering approaches are typically not user-centered and consequently are often the cause of a bottleneck in system development. This paper describes and evaluates the Advanced Knowledge And Design Acquisition Methodology (AKADAM) project as an attempt to overcome such inadequacies by provision of user-centered knowledge acquisition techniques. Both theoretical and practical issues are examined. The role of multiple perspectives (i.e. "knowledge as rules", "knowledge as concepts", and "knowledge as designs"), their relationship to a user-centered approach, and the necessity of flexible knowledge integration are portrayed by applying AKADAM to a complex, real-world domain (i.e. the development of an electronic associate for fighter pilots). Results suggest that this approach is capable of providing: (a) a naturalistic knowledge elicitation environment endorsed by users, (b) an externalization of experts' intuitive knowledge in a form which is similar to their own mental representation and (c) an integrated, large-scale knowledge set suitable for infusing knowledge into AI architectures and human-computer interface design.

This paper presents an application of knowledge intensive generalization to knowledge acquisition, in the domain of air traffic control. We explain why knowledge intensiveness and first-order logic are sometimes necessary, as for instance in the application field studied here. An obvious advantage of first-order logic is its power of expression, while an obvious drawback is long computation time. We also describe some less obvious advantages and drawbacks of first-order logic, especially when the knowledge must be expressed as Horn clauses to retain some computational efficiency. Finally, we emphasize the large translation problem that must be solved in order to allow an efficient interaction with the expert. Two translation phases are necessary. One goes from the expert's language to Horn clauses, the second one goes back from Horn clauses to the expert's language. The first one is necessary to ensure automatic learning, while the second one allows the expert to understand what has been learned. Both phases are far from trivial and ask for choices that must be made carefully in order to avoid losing significant information. One of our unexpected results is that the second translation phase plays the role of a validation step. It thus becomes a very efficient way to acquire knowledge the expert has problems formalizing. Using first-order logic does complicate things, but it provides, as a reward, a powerful way of extracting and validating the acquired knowledge, especially when the field expert is unable to express his knowledge in a simple way.