Knowledge Acquisition最新文献

Inductive learning algorithms have been suggested as alternatives to knowledge acquisition for expert systems. However, the application of machine learning algorithms often involves a number of subsidiary tasks to be performed as well as algorithm execution itself. It is important to help the domain expert manipulate his or her data so they are suitable for a specific algorithm, and subsequently to assess the algorithm results. These activities are often called preprocessing and postprocessing. This paper discusses issues related to the application of the ID3 algorithm, an important representative of the inductive learning family. A prototype workbench which has been developed to provide an integrated approach to the application of ID3 is presented. The design rationale and the potential use of the system is justified. Finally, future directions and further enhancements of the workbench are discussed.

Existing group decision support systems used in meeting rooms can help teams reach decisions quickly and efficiently. However, the decision models used by these systems are inadequate for many types of problems. This paper describes our laboratory's experience with knowledge acquisition systems and decision support tools. Our studies led us to develop a comprehensive decision model for group decision support systems. This decision model combines current brainstorming-oriented methods, structured text argumentation (using the gIBIS model), repertory grids, possibility tables (morphological charts) and influence diagrams from decision analysis. Each component addresses weaknesses in current group decision support systems. We are assembling these group decision support components together into a group decision workbench.

One aspect of knowledge acquisition involves reaching an understanding of how an expert combines certainties in the course of reasoning. There are several distinct junctures in an inference process where certainties need to be combined. At the most elemental level, these include combining certainties of operands involved in the arithmetic, logical and relational expressions that can constitute a premise. As a basic frame of reference for acquiring knowledge about certainty treatments, there is a prescriptive mapping of operators into the joint and confirmative classes of certainty factor algebras. However, these prescriptions have not been empirically studied. Here, we report on an experiment conducted to test various hypotheses about actual behaviors of people in combining certainties for elemental operators. For each operator, we found behaviors to be consistent with prescriptions in some respects, but deviating from them in other respects. The result is an empirical base from which to launch efforts involving the acquisition of knowledge about how specific experts combine certainties at various junctures in inference processes.

This paper contains a philosophical examination of the generic task methodology as developed by B. Chandrasekaran and others. Two phases in the evolution of this methodology are discerned. The earlier, "Platonic" phase resulted in a methodology in which the notions of "task" and "method" were very closely coupled. This led to a tension between two functions of generic tasks: the conceptualization of a task would ipso facto include some commitment to an AI method, but typically, the criteria for a task analysis are different from those for choosing an AI method. In the later phase of the generic task methodology, a generic task is to be seen as an analysis of a task, issuing in a task structure. The connection between tasks and the methods for their performance is loosened, but not severed. This entails that the same philosophical problems re-emerge, albeit in a less virulent form. If the task structure is to be seen as an analysis of the task, then that impairs its function as an AI methodology, and vice versa. This paper concludes with the setting out of a thoroughgoing anti-realist philosophy of mind which enables the generic task view to avoid many of these problems.

We describe a general purpose knowledge management system, discussing its general goals and features, as well as its use in several very different applications. By "multi-functional", we mean having a wide variety of knowledge management functions such as debugging, formatting, and retrieval, and a wide variety of possible applications. The system, called CODE, functions primarily as a "knowledge engineer's rapid prototyper", or as a "spreadsheet for ideas"; one can experiment rapidly with relationships between concepts and obtain quick feedback on the desirability of changes and additions to a knowledge base. CODE's highly graphic interface permits experimentation with descriptions or definitions of concepts, which are arranged in an inheritance network using a very flexible inheritance mechanism. Several associated subsystems, such as a first order logic system and a simple natural language system, allow various types of syntactic and semantic checks to be performed if desired. We illustrate CODE's flexibility by describing three typical applications: in software engineering, terminology, and ontological design for knowledge-based systems.

Recent exploratory research developed and tested a guided self-elicitation (GSE) methodology. With GSE, an expert is enabled to capture his/her own performed expertise as production rule-instances. GSE is based on published cognitive research, using a production system view of conscious cognitive information processing and certain demonstrated human abilities: for identifying and categorizing perception, rehearsing and reconstituting prior thought processes and verbal reporting of concurrent cognitive information processing. Experimentally self-elicited decision analyst expertise (leading subjective probability assessment interviews) demonstrates that performed expertise can consist of complex rule-processed knowledge forms. An object model for representing complex knowledge forms is proposed and discussed.

Building knowledge-based problem solvers requires an intellectually challenging modeling stage whose dominance over other activities is now widely recognized. In spite of this, current languages and environments leave the modeling activity on the shoulders of the human, concentrating on the routine programming aspect. Next generation languages and tools will have to explicitly support modeling in the first place. This paper presents a proposal for such a next generation knowledge modeling environment and discusses some steps we have made in this direction. Unlike existing programming environments, knowledge modeling environments focus on manipulating explicit, declarative specifications of problem-solving models which must be acquired, organized, modified, explained, validated, simulated and, eventually, translated into performance computer languages. Programming is only one of the activities supported in such an environment. This paper also discusses the knowledge modeling language we have developed as the foundation of the modeling environment. This language extends term classification technology with refinement, constraints, patterns and events, actions and methods, in order to support the description of both domain and control specifications required by problem-solving models. To substantiate the claims about the adequacy of the language, the paper presents two important modeling applications. The first is developing a full KADS language on top of it and the second is modeling a well known generic problem solving method, "propose-and-revise".