Such complex concepts as “thinking” and “consciousness”, and even more easily defined ones, such as “intelligence” and “knowledge”, among specialists of various profiles (for example, systems analysis, computer science, neuropsychology, psychology, philosophy, etc.) can differ significantly.
Complete, adequate representation of knowledge, which is perceived equally unambiguously by both people and machines, is the main problem of modern information exchange. Such information exchange is based on a system of concepts and relationships that make up knowledge.
Classification of knowledge
They can be classified into several categories: conceptual, constructive, procedural, factual and metaknowledge.
- Conceptual knowledge is a set of specific concepts used in solving problems. They are often used in the fundamental sciences and theoretical fields of science. In fact, conceptual knowledge constitutes the conceptual apparatus of science.
- Constructive knowledge - sets of structures, systems and subsystems, as well asinteractions between them. Actively used in technology.
- Procedural knowledge is the methods and algorithms most commonly used in applied sciences.
- Factual knowledge is the characteristics of objects and phenomena, both quantitative and qualitative. Most often used in experimental sciences.
- Metaknowledge is any knowledge about knowledge, its system of organization, its engineering, and about the order and rules of its application.
Knowledge organization
Knowledge organization system is the process of providing information in the form of messages that can be familiar (oral and written speech, pictures, etc.) and unusual (formulas, map objects, radio waves, etc.).
For a knowledge organization system to be understandable and successful, it is necessary to use an understandable and constructive system of rules according to which knowledge will be presented and perceived. To do this, a person uses language and writing.
Language
Language appeared and developed due to the fact that the knowledge accumulated by people constantly needs to be presented, expressed, stored and exchanged. A thought that cannot be expressed by a formal construction (language, image) loses the opportunity to become part of the information exchange. That is why throughout the history of mankind, language has been the most effective form of knowledge representation.
The richer the language, the more knowledge it expresses, respectively, making the culture of the people richer, which, in turn, allows you to develop more and more effective systems of organizing knowledge.
Languagescience
The main problem in using language as a form of knowledge representation is the ambiguous semantic meaning of words and sentences. That is why the language of science plays a special role in the formalization of knowledge.
The main purpose of the language of science is to typify and standardize the forms of expression, compression and storage of knowledge. With the help of a typical, standard presentation of knowledge, one can get rid of polysemy or semantic ambiguity of the language.
What, in the natural conditions of language evolution, makes a language richer (polysemy of expressions), becomes a hindrance in the process of knowledge exchange, increasing the risk of misunderstanding, semantic noise and ambiguous perception of information.
Classification of knowledge
One of the main methods of knowledge formalization is classification. This is the distribution of knowledge into groups in accordance with a certain class. That is, only information that meets certain criteria corresponding to the class falls into a certain class of knowledge.
Classification is a particularly important method of scientific systematics, which is indispensable at the first stage of the formation of basic knowledge of a scientific direction. For example, in computer science without classification, there is no equivalence that allows you to solve such important tasks as comparison, search and categorization. Without classification in science, we would not have such unique and invaluable data organization systems as the periodic table.
Knowledge Representation Models
The periodic table, the Table of Ranks, the Criminal Code, family trees and other classification systems are models of knowledge representation. These are formal structures that link certain knowledge: facts, phenomena, concepts, processes, objects, relationships.
To understand and process knowledge about a particular subject area by a computer, this knowledge must be presented in a certain, formalized form. Depending on the purpose, the processing of knowledge by a computer occurs in accordance with a model built on an algorithm. Accordingly, the knowledge presented in the model depends on the algorithm for processing it.
There are several models of knowledge representation in expert systems. The main ones are production, frame, network and logical.
Classification of models
The knowledge representation models listed above, examples of which follow, although widespread, are far from the only ones. Today, there are many models that differ from each other in terms of validity, approaches to their creation and organization principles.
For example, the table below shows the types of knowledge representation models, their division into empirical and theoretical, as well as further subdivision.
| Empirical models | Theoretical models |
| Production models | Logical models |
| Network models | Formal grammars |
| Frame models | Combinatorial models |
| Lenemy | Algebraic models |
| Neural networks | |
| Genetic algorithms |
Empirical modeling
Empirical models of organization and representation of knowledge take a person as an example and try to embody the organization of his memory, consciousness and decision-making and problem-solving mechanisms. Empirical modeling refers to any kind of model built on the basis of empirical observations, rather than relationships that can be mathematically described and modelled.
Empirical modeling is a general term for knowledge representation models that are created on the basis of observations and experiments.
An empirical model operates according to a simple semantic principle: the creator observes the interaction of the model and its referent. The processing of information received can be "empirical" in many ways, from analytical formulas, causal relationships, to trial and error.
Production models of knowledge representation
This data representation model is most often based on relationships and causality. If the information can be represented in the form of conditions of the type "If, Then", then the model is production. It is most often used in applications and simple artificialintelligence.
Knowledge representation production models are most often computer programs that provide some form of artificial intelligence with a set of rules of behavior, as well as the mechanism necessary to follow these rules under certain conditions.
Production (a set of rules) consists of two parts: a precondition ("IF") and an action ("THEN"). If the production precondition matches the current state of the world, then the model runs. The production model also contains a database, sometimes referred to as working memory, which contains current knowledge.
The disadvantages of the production model are that if the number of rules is too large, the actions of the model may contradict each other.
Semantic networks
They are based on the integrity of the image and are the most visual models of knowledge representation. The semantic network is most often represented as a graph or a complex graph structure, the nodes or vertices of which represent objects, concepts, phenomena, and the edges represent relationships between certain objects, concepts and phenomena.
The simplest semantic network can be easily represented as a triangle, the vertices of which are such concepts as, say, "dog", "mammal" and "spine". In this case, the vertices will connect the sides of the triangle, which can be denoted by such connections and relationships as "is", "possesses", "has". in this way we get a knowledge representation model from which we learn,that a dog is a mammal, mammals have a backbone, and a dog has a backbone.
Such models are illustrative, and with their help, you can most effectively represent complex systems and causal relationships. In addition, these semantic networks can be supplemented with new knowledge by expanding an existing network, that is, a triangle can be turned into a rectangle, then into a hexagon, and then into a complex network of intersecting shapes, in which one can observe, for example, inheritance of properties.
Frame model
The frame model is named so from the English word frame - frame or frame. A frame is a structure that collects data used to represent a particular concept.
As in sociology, where frames are a kind of stereotyped data that influences the human perception of the world and the decision-making process, in computer science and working with artificial intelligence, frames are used to create structured data that represents stereotypical situations. In fact, this is the initial, basic data system on which the perception of the world by artificial intelligence is built.
Besides being effective models of knowledge representation, frames are active not only in computer science. They were originally a variation of semantic networks.
A frame consists of one or more slots. In turn, slots can themselves be frames. Thus, the frame model is able to represent complex conceptual objects, forming a wide hierarchical chain.knowledge.
The knowledge representation frame model contains information about how to use a frame, what to expect during and after using it, and what to do when expectations from using a frame are not met.
Certain kinds of data in a frame model are fixed, while other data, usually stored in terminal slots, can change. Terminal slots are most often treated as variables. Top-level slots and frames carry information about the situation, which is always true, but terminal slots do not have to be true.
Frames of one complex network can share the slots of other frames of the same network.
The database can store prototype frames (immutable) and instance frames that are created situationally to represent a particular situation or concept.
Frame models of knowledge representation are one of the most versatile and capable of displaying various types of knowledge:
- frame structures are used to represent concepts and objects;
- frame roles denote role responsibilities;
- frame scripts describe behavior;
- frame situations are used to represent state and activities.
Neural networks
These algorithms can also be conditionally added to the group of models based on an empirical approach to knowledge. In fact, neural networks are trying to copy the processes occurring in the human brain. They are based on the theory that an artificial intelligence system with the same structures andprocesses, as in the human brain, will be able to get similar results in the process of decision-making, evaluation of situations and perception of reality.
Theoretically sound approach
Mathematical, predicative and logical models of knowledge representation are based on this approach. These models guarantee the correctness of decisions because they are based on formal logic. They are suitable for solving simple problems from a narrow subject area, often associated with formal logic.
Logical models of knowledge representation
This is one of the most popular models based on a theoretical approach. The logical model uses the predicate algebra, its system of axioms and inference rules. The most common logical models use terms - logical constants, functions and variables, as well as predicates, that is, expressions of logical actions.
