Expert systems (representation of knowledge)
Idea of knowledge in expert system of the problem environment needs to be implemented so that the expert system could manage process of search of the solution, was capable to acquire new knowledge and to explain the actions. It should be able to work not only with knowledge, but also over knowledge i.e. to have capability to understand and investigate them. Thus, the expert system should have knowledge of how her knowledge of the problem sphere is provided.
Content |
Presentation layers
If to call knowledge of the problem environment knowledge of zero level of representation, then the first presentation layer contains metaknowledge, i.e. knowledge of how are provided in inner world of a system of knowledge of zero level. The first level contains knowledge of those means which is used for representation of knowledge at zero level. Knowledge of the first level has important appointment in process control of the solution, in acquisition of new knowledge and an explanation of actions of a system. Knowledge of the first level does not depend on the problem environment for the reason that they do not contain references to knowledge of zero level. The number of the presentation layers can be and more than two. The second presentation layer will contain in this case data on knowledge of the first level, i.e. knowledge of representation of basic concepts of the first level. Separation of knowledge of the presentation layers provides expansion of area of applicability of a system.
Detail levels
Introduction of levels of detail allows to consider knowledge with different degree of a detail. The number of levels of detail in many respects depends on specifics of solvable tasks, volume of knowledge and a method of their representation. Usually, create not less than three levels of detail by which the general, logical and physical organizations of knowledge are defined respectively. Selection of several levels of detail gives additional degree of flexibility of a system as allows to make changes at one level, without affecting others. Changes at one level of detail can lead to additional changes at the same level that occasionally is necessary for ensuring coordination of data structures and programs. However presence of different levels interferes with distribution of changes from one level on others.
The organization of knowledge in a working system
The working storage of expert systems is the database (see. Expert systems (architecture)). Data can be provided in a working storage homogeneous or be separated into levels on the types. Data of the corresponding type are stored in the second case at each level of a working storage. Separation into levels complicates structure of expert system, but tells her big efficiency. For example, it is possible to select the level of plans, level of the ordered list of the rules ready to accomplishment and level of solutions. In modern expert systems data in the working storage (WS) can be isolated or connected. In the first case the working storage consists of a set of simple elements, in the second - of one or several difficult elements. At the same time the difficult element corresponds to a set of the integrated simple. Theoretically both approaches provide completeness, but use of the isolated elements in difficult data domains leads to loss of efficiency. Data in RP in the elementary can be constant, variable or mixed. At the same time variables can be considered as characteristics of a certain object, and a constant – as values of the corresponding characteristics. If in RP it is required to analyze at the same time several different objects describing the developed problem situation, then it is necessary to specify to what objects the considered characteristics belong. One of methods of the solution of this task are the explicit indication of to what object characteristic belongs.
If the working storage consists of difficult elements, then communication between separate objects is specified obviously, for example a task of semantic relations. At the same time each object can have the inner pattern. For acceleration of search and comparison data in RP can be connected not only logically, but also is associative.
The organization of knowledge in the database
Criterion of intellectuality of a system in terms of representation of knowledge the ability of a system to use at the right time necessary (relevant) knowledge is considered. Systems which do not have means of the analysis of relevant knowledge inevitably face the problem called by "combinatorial explosion". This problem is one of the basic reasons limiting scope of expert systems. In a problem of access to knowledge it is possible to select three aspects: connectivity (or aggregation) knowledge and data, an access mechanism to knowledge and a comparison method.
Connectivity of knowledge
The coherence of knowledge is the main method providing fall forward of search of relevant data. Popular belief that knowledge should be organized around the most important objects of data domain. All knowledge characterizing some entity communicates and presented in the form of a separate object. At the similar organization of knowledge if information on some entity is necessary for a system, then she looks for the object describing this entity, and then in an object finds information on this entity. In objects select two types of sheaves between elements: external and internal. Internal sheaves will organize elements in a uniform object and express structure of an object. External sheaves indicate interrelations, between objects in the field of examination. There is a division of external sheaves into logical and associative. Logical reflect semantic relations between elements of knowledge. Associative sheaves provide the interrelations promoting fall forward of process of search of relevant knowledge.
Problem of search of knowledge
The main problem during the work with the big knowledge base is the problem of search of knowledge relating to a solvable task. Because in the processed data it can not be contained explicit references to the values required for their processing certain more general search mechanism, than the direct access method (a method of explicit links) is necessary. This method is intended in order that on some measure description which is available in a working storage to find already in the knowledge base the objects satisfying to this description. Thus, streamlining and structuring knowledge can accelerate search process considerably. Data retrieval can be considered as the process consisting of two stages. The first stage corresponds to selection process on associative sheaves. There is preselection of potential candidates for a role of desirable objects in the knowledge base. At the second stage by accomplishment of transaction of comparison of potential candidates to descriptions of candidates the final choice of required objects is performed Comparison transaction can be also used for classification, confirmation, decomposition and correction. For identification of an unknown object it can be compared with some known samples. It will allow to classify an unknown object as such known sample by comparison to which the best results were received. Comparison is used for confirmation of some candidates from a set of possible. If to compare some known object with the unknown description, then in case of successful comparison partial decomposition of the description will be made.
Finding solutions methods
The methods of solving of tasks based on data of these tasks to search depend on features of data domain in which the problem and from requirements imposed to the solution is solved. Features of a problem area are defined by the following parameters:
- solution search space volume
- convertibility of area in time and space
- completeness of the model describing area. If the model is not complete, then for area declaration several models supplementing each other are used.
- determinancy of data on a solvable task
User requirements to result of the task solved using search can be determined by the number of solutions and properties of result and (or) method of its receiving. So, the task can have one solution, several solutions, all solutions. Properties of the solution set restrictions with which should satisfy the received result or a method of its receiving. For example, for the system issuing recommendations to repair of machines, the user can specify the requirement not to use a part of some producers of a price category or properties of metals. The properties parameter can define also such features as time allowed for the solution, amount of memory, occupied for obtaining result, the instruction on obligation of use of any knowledge, etc.
The complexity of the task determined by such parameters varies from simple problems of small dimension with unchangeable certain data and lack of restrictions for result with method of its receiving to difficult tasks of big dimension with impure, wrong and incomplete data and any restrictions for result and a method of its receiving. By any one method it is impossible to solve all problems. Usually some methods exceed others only by some of the listed parameters.
The existing methods the solving of tasks used in expert systems can be classified as follows:
- search methods in one space (area of small dimension, completeness of model, exact and complete data)
- search methods in hierarchical spaces (areas of big dimension)
- search methods at inexact and incomplete data
- the search methods using several models
These methods if necessary should integrate to allow to solve problems which complexity increases at the same time by several parameters.
Connected subjects
Links
| The site content is translated by machine translation software powered by PROMT. The machine-translated articles are not always perfect and may contain errors in vocabulary, syntax or grammar. Read original article If you find inaccuracies or errors in the results of machine translation, please write to editor@tadviser.ru. We will make every effort to correct them as soon as possible. |