![]() |
|||
![]()
|
![]() |
![]() |
![]() |
2.3. PATTERNSMost of the everyday situations are hardly to be represented by logic and the usual ways of uncertainty calculations. As it was mentioned in the introduction, we somehow receive and store coherent impressions that we name patterns, using the visual metaphor. The origin of the pattern idea can be traced back to Plato and further. It always returned at the encounters with the limits of rational thinking, i.e., logic. Scientific investigation started in psychology. In the late 19th century, Ehrenfels coined the German name of shape, appearance, Gestalt, for these pattern-related phenomena, yet the field is still investigated mainly by psychology; scheme is one of the favorite namings. Cognitive psychology, the psychology of cognition, knowledge acquisition, and brain processing, is a close companion of computer scientists and designers of expert systems. Mathematical tools for identification, separation, similarity measures of patterns come mainly from statistics, applications of geometrical metaphors (multidimensional, differently shaped, solid objects) and based on that, linear algebra, which is still the most useful tool for separating components of a complex phenomenon. 3. TECHNIQUES, PRACTICES, AND METHODOLOGIES3.1. CONCEPTUAL REPRESENTATIONConcept is a generalization of certain directly experienced objects, phenomena, situations, events. Though Platonistic philosophers believe in the independent existence of concepts (ideas), expert systems cannot apply these metaphysical beliefs; computers which should finally represent some kind of knowledge can start only from the basics, to do anything practical. This means that representation of any kind of knowledge starts with learning and programming of conceptual structures, an attempt to find hierarchies of the subject concerned. Having experienced individual apples and oranges, the concept of a general apple and orange can be built (defined), thereafter the concept of fruits, food, etc. The same way is valid for the conceptualization of body-related phenomena of medical expertise, commodity-related concepts of economy, elementary geometry, material and functional parts of engineering, basic structures of spatial delimitation in architecture, tangible damage in legal practice. Knowledge engineering, a special branch of expert system design, evokes and helps to learn these conceptual structures inquiring domain experts and other information resources. Each object is characterized by its attributes, like shape, color, composition, etc. and measures (values) of these attributes, qualitative (large, strong, opaque, full) and quantitative in certain well-defined and basic object-originated dimension (like the meter-bar in Paris, or an atomic clock). In building conceptual hierarchy the common features of these characteristics should be accumulated and preserved because the main use of concepts is to provide an easy reference to a greater number of objects, phenomena under the hierarchical structure of the concept. The singular entity should inherit these characteristics from the concept and vice versa. Conceptualization is, on the one hand, a shorthand for characterization; on the other hand, a basis for reasoning: the first hypothesis for a decision, action is the idea that the same, or at least similar procedure can yield similar effects within the range of a whole conceptual bunch. This latter virtue of conceptual representation hints at the practice of attaching not only attributes and values, but also usual procedures to the representation scheme, like certain checks, medicines, therapies to a certain malady defined on a conceptual level, or a spell checker to the concept of a practical word processor. The conceptual structure is not too different from any database system, especially not from those sold as intelligent systems, relational databases. Structures programmed in the object-oriented sense bear similar characteristics; for the most part, some of the names of the ingredients are different. In the object-oriented vernacular the object is the bottom concept, that one which is not differentiated further within the system. This means that the chain of definitions and reasoning based on that can start with one particular apple, with the general apple concept or with fruit, according to users' requirements. The next levels of objects are classes, within the classes the particular objects are instances. The classes and objects have describing, definitional variables. Those which characterize a class are the vehicles of the mentioned inheritance procedure, one of the keys for reasoning. Objects are embedded into classes, some closer groups of objects into subclasses, several coherent classes are embedded in superclasses. All of these represent the conceptual hierarchies. For explanation, as used in dictionaries, other terms are used; these are the metaclasses. A class or an object can call a procedure, e.g., the program of peeling an apple, or in a general way, any fruit. These are the methods. The call is a message, and these messages are chosen and combined by selectors (Figure 1). All these are based on the ancient and most simple logical formulae, an object-class relation is defined by the is_a junction, the list of definition attributes is collected by practical viewpoints but not very different from those used more than 2000 years ago: quantity, quality, place, time, situation, possession, action, etc. In a more advanced form, this structure is a frame; within this frame name, attributes and other ingredients have prefixed places, slots; the slot place itself defines the route of the input and of the output, i.e., the interpretation, application of the content, makes the user's work easier. Keys, used in relational databases, are compositions of attributes that collect the same or similar objects from a larger data (knowledge) base, according to a certain practical objective, e.g., collecting all fruits with removable skin and core.
This very simple and obvious structure of knowledge-database is an essential part of knowledge representation. The accomplishment of such a database in any real life, for a larger scale task is a tremendous professional work and responsibility. Beyond all the usual requirements of large databases, like consistency, possibility and control of multiple use, data integrity, the knowledge base should handle and convert verbal and numerical information. As the most important feature, as emphasized, it should be the basic material for further reasoning. This means that the structure must be available for all further complex reasoning operations, open for appropriate and closed for not applicable reuse, and it must provide a fast, efficient, highly reliable interface both for the user and the inference machine. The conceptual inheritance is a powerful tool for reasoning but multiple embedding always carries the danger of conflicting and not applicable consequences. The user interface should have attractive, friendly access for non-computer-oriented users, and be an active, stimulating device for knowledge acquisition. Due to these requirements, only professional, well-documented and referenced, supported knowledge base tools are advisable for real-life applications. A careful selection, possibly some previous and extensive tests of the system, are recommended, whether or not it is the best match for the task concerned.
|
![]() |
|
Use of this site is subject certain Terms & Conditions. Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details. |