Knowledge engineering and knowledge acquisition. Knowledge Engineering Knowledge Representation Models

An engineering discipline that deals with the integration of knowledge with computer systems to solve complex problems that typically require a high level of human expertise:

knowledge configuration management (accounting);
change management (evolution);
logistics (search and delivery as needed).

At a high level, the knowledge engineering process consists of two:

Knowledge extraction- transformation of “raw knowledge” into organized knowledge, the process of obtaining knowledge from its sources, which can be material media (files, documents, books) and experts (groups of experts). It is part of Knowledge Engineering.
Implementation of knowledge- transformation of organized knowledge into realized, the process of transforming organized knowledge into realized.

Knowledge Management Technologies

The following knowledge management technologies are distinguished:

working with tacit knowledge(tacit knowledge) in the minds of experts(most often they are what is meant when we talk about “knowledge management”). Cognitologist (role):
- helps the expert identify and structure the knowledge necessary for the operation of the expert system, extracts informal knowledge from the expert;
- selects the intellectual system that is most suitable for a given problem area, and determines the way knowledge is represented in this IS;
- selects and programs standard functions that will be used in the rules entered by the expert.
working with written knowledge("knowledge management" extends to computers: corporate knowledge management, Knowledge Management) - emphasis on "full-text search", "semantic search", "automatic annotation".
1. NLP as a datalogical discipline (“work according to form”), wave technique, perceptual modalities, submodalities, spatial marking, calibration
2. use of web 2.0 (blogs and wikis)
working with written formal knowledge(knowledge engineering, which is also included in knowledge management, but not so confidently) - emphasis on structural databases, engineering models, data integration. Most technologies in knowledge engineering have followed the path of implementing the so-called “semantic network”, the Husserl-Wittgenstein-Bunge approach that knowledge is representable by facts (and facts are relationships between concepts). From a set of facts, a semantic network arises (see review by John F. Sowa), in which edge relations connect vertex concepts. The idea of storing and using knowledge in semantic form was implemented by many almost non-overlapping communities of practice, which resulted in the emergence of a huge number of implementations and standards in which not a single word is the same, but which are ideologically and technologically compatible.
1. Data Modeling + Data Integration. They are used when it is necessary to combine data from multiple CAD systems from different suppliers during the construction of a large industrial facility. Keywords: ISO 15926, Gellish, ISO 10303. Instead of the word "ontology" they say "data model". : practically none, all data queries. Everyone fights with knowledge hand-to-hand. No graphics, all XML, proprietary storage formats Data schemas in each individual CAD system. Recently, other solutions have appeared aimed at integrating heterogeneous data, for example from CYC and (based on the standardized UMBEL ontology, expression in RDF and providing access to data via HTTP, see). ISO 15926-7 projects come down to the same thing: a certain ontology + semantic web standards.
2. Concept Map() Used for (often collaborative via the web) educational and creative work. Key formats (all in XML): XCT 3.0, but ready to eat Topic Map, and much more for editing and display. Knowledge management tools: graphical display, combining networks that were drawn by two participants in the creative process. A close relative is MindMap, where there is not a graph at all, but a beautifully drawn tree, and the connections are unnamed.
3. Conceptual Graphs Used for academic studies in artificial intelligence, expert systems, agent systems and other classics of the genre. They rely on the work of the philosopher and logician Pierce ("intelligent indexing"), the key person is John F. Sowa. Key format for storing knowledge: three syntaxes, the main one being CGIF (XML). Knowledge management tools: Common Logic (or ISO ISO/IEC IS 24707:2007, ).
4. Topic Map They will be used for Knowledge Management initiatives - and they came from catalogers (bibliographers). Big fans of standardization (see), but have lost focus (they are inexorably drawn to general data modeling, in which they lose to Semantic Web approaches). Key formats for storing knowledge: ISO 13250, XTM 2.0, HyTM. Knowledge management tools: topic map engine is used (dozen options), because TMAPI 2.0 is standardized. In addition, a special standard for specifying constraints for topic maps - ISO/IEC FCD 19756 (TMCL) - has reached the finish line, and the Topic Map Query Language (ISO 18048 project) seems to have stalled.

17.2. Practical knowledge extraction methods

17.3. Structuring knowledge

The central problem in creating intelligent information technologies is the adequate representation of a specialist’s knowledge in computer memory. This led to the development of a new direction in computer science - knowledge engineering, where the relationship between human knowledge and its formalized (information) display in a computer is determined. Knowledge engineering studies and develops issues related to the acquisition of knowledge, its analysis and formalization for further implementation in an intelligent system.

Purpose of the chapter– give an overview of the main theoretical aspects of knowledge engineering and introduce some practical methods of working as knowledge engineers.

After studying this chapter you should know:

Approaches to acquiring knowledge when developing expert systems

Theoretical problems arising in knowledge retrieval

Features of psychological and linguistic factors that need to be taken into account by a knowledge engineer

The influence of the philosophy of knowledge on the work of a knowledge engineer

Knowledge engineer methods when working with a knowledge source

Knowledge extraction methods

The essence of expert games

Methods for extracting knowledge from texts

Structuring the acquired knowledge

Formation of the conceptual and functional structure of the subject area

How is knowledge formalized and a knowledge base formed?

17.1. Theoretical aspects of acquiring knowledge

Strategies for acquiring knowledge

Psychological aspect

Linguistic aspect

Epistemological aspect

STRATEGIES FOR OBTAINING KNOWLEDGE

There are several strategies for acquiring knowledge. The most common:

acquisition;

extraction;

formation.

Under acquisition of knowledge understands the method of automated construction of a knowledge base through a dialogue between an expert and a special program (in this case, the structure of the knowledge is pre-built into the program). This strategy requires significant preliminary study of the subject area. Knowledge acquisition systems actually acquire ready-made pieces of knowledge in accordance with the structures laid down by the system developers. Most of these tools are specifically focused on specific expert systems with a strictly defined subject area and knowledge representation model, i.e. are not universal. For example, the TEIRESIAS system, which became the progenitor of all tools for acquiring knowledge, is intended to replenish the knowledge base of the MYCIN system or its subsidiary branches, built on the EMYCIN “shell” in the field of medical diagnostics using a production model representation knowledge.

Term knowledge extraction concerns direct live contact between the knowledge engineer and the knowledge source. The authors tend to use this term as a more capacious and more accurately expressive meaning of the procedure for transferring the competence of an expert through a knowledge engineer into the knowledge base of an expert system.

Term formsupknowledge acquisition has traditionally been assigned to the extremely promising and actively developing field of knowledge engineering, which deals with the development of models, methods and algorithms for data analysis for knowledge acquisition and learning. This area includes inductive models for generating hypotheses from training samples, learning by analogy, and other methods.

Thus, we can distinguish three strategies for conducting the knowledge acquisition stage in the development of expert systems (Fig. 17.1).

Rice. 17.1. Three Strategies for Gaining Knowledge

At the present stage of development of expert systems in our country, the strategy of knowledge extraction is apparently the most relevant, since there are practically no industrial systems for acquiring and forming knowledge in the domestic software market.

Knowledge extraction– this is a procedure for interaction between an expert and a source of knowledge, as a result of which the reasoning process of specialists when making a decision and the structure of their ideas about the subject area become explicit.

Currently, most developers of expert systems note that the process of knowledge extraction remains the biggest bottleneck in the construction of industrial systems.

The process of knowledge extraction is a long and labor-intensive procedure in which a knowledge engineer, armed with special knowledge in cognitive psychology, systems analysis, mathematical logic, etc., needs to recreate a model of the subject area that experts use to make decisions. Often, beginning developers of expert systems, wanting to avoid this painful procedure, ask the question: can an expert extract knowledge from himself? For many reasons this is undesirable.

Firstly, most of an expert’s knowledge is the result of numerous layers, stages of experience. And often knowing that from A should IN, the expert does not realize that his chain of reasoning was much longer, for example WITH D, D  A, A IN, or A Q, Q  R, R  B.

Secondly, as the ancients knew (remember Plato’s “Dialogues”), thinking is dialogical. And therefore, the dialogue between a knowledge engineer and an expert is the most natural form of “unwinding” the labyrinths of the expert’s memory, in which knowledge is stored, partly of a non-verbal nature, i.e. expressed not in the form of words, in the form of visual images, for example. It is in the process of explaining to the knowledge engineer that the expert puts clear verbal labels on these blurred associative images, i.e. verbalizes knowledge.

Thirdly, it is much more difficult for an expert to create a model of the subject area due to the depth and vastness of the information that he possesses. Numerous cause-and-effect relationships of a real subject area form a complex system, from which isolating the “skeleton”, or the main structure, is sometimes more accessible to an analyst who also owns a systematic methodology: Any model is a simplification, and it is easier to simplify with less knowledge of the details.

To understand the nature of knowledge extraction, we highlight three main aspects of this procedure (Fig. 17.2): psychological, linguistic, epistemological, which are described in detail in.

Rice. 17.2. Key Aspects of Knowledge Retrieval

PSYCHOLOGICAL ASPECT

Communication model for knowledge retrieval

Of the three identified aspects of knowledge extraction psychological is, apparently, the main thing, since it determines the success and efficiency of interaction between a knowledge engineer (analyst) and the main source of knowledge - a professional expert. We highlight the psychological aspect also because knowledge extraction most often occurs in the process of direct communication between system developers.

The desire and ability to communicate can characterize the degree of professionalism of a knowledge engineer.

It is known that information losses during conversational communication are large (Fig. 17.3). In this regard, we will consider the problem of increasing the information content of communication between an analyst and an expert through the use of psychological knowledge.

Rice. 17.3. Loss of information during communication

We can propose the following structural model of communication when extracting knowledge:

communication participants (partners);

means of communication (procedure);

subject of communication (knowledge).

In accordance with this structure, we will highlight three “layers” of psychological problems that arise when extracting knowledge (Fig. 17.4), and consider them sequentially.

Rice. 17.4. The structure of the psychological aspect of knowledge extraction

contact layer

Almost all psychologists note that any collective process is influenced by the atmosphere that arises in the group of participants. There are experiments whose results undeniably show that the friendly atmosphere in a team has a greater influence on the result than the individual abilities of individual group members. It is especially important that the development team develops cooperative rather than competitive relationships. Cooperation is characterized by an atmosphere of cooperation, mutual assistance, interest in each other’s successes, i.e. the level of moral communication, and for competitive relationships - an atmosphere of individualism and interpersonal rivalry (lower level of communication).

Unfortunately, it is impossible to predict compatibility in communication with a 100% guarantee. However, it is possible to identify a number of personality traits, character and other characteristics of the participants in communication, which undoubtedly influence the effectiveness of the procedure. Knowledge of these psychological patterns is part of the psychological culture that a knowledge engineer must have to successfully carry out the knowledge extraction stage:

goodwill and friendliness;

sense of humor;

good memory and attention;

observation;

imagination and impressionability;

greater concentration and perseverance;

sociability and resourcefulness;

analyticity;

attractive appearance and manner of dressing;

self confidence.

Procedural layer

A knowledge engineer who has successfully mastered the science of trust and rapport with an expert (the contact layer) must still be able to take advantage of the beneficial effects of this atmosphere. The problems of the procedural layer concern the knowledge extraction procedure itself. There is little insight and charm useful for solving the problem of contact; professional knowledge is required here.

Let us dwell on the general principles of the procedure.

A conversation with an expert is best conducted in a small tête-à-tête room. Lighting, warmth, comfort directly affect your mood. Tea or coffee will create a friendly atmosphere. American psychologist I. Atwater believes that for business communication the most favorable distance is from 1.2 to 3 m. The minimum “comfortable” distance can be considered 0.7 - 0.8 m.

Reconstructing your own reasoning is not easy work, and therefore the duration of one session usually does not exceed 1.5 - 2 hours. It is better to choose these two hours in the first half of the day (for example, from 10 to 12 o'clock). It is known that mutual fatigue of partners during a conversation usually occurs after 20 - 25 minutes, so pauses are needed in the session.

Every knowledge engineer has his own unique way of speaking. Some speak quickly, others speak slowly; some are loud, others are quiet, etc. It is almost impossible to change the style of conversation - it is ingrained in a person in early childhood. However, knowledge extraction is a professional conversation, and its success is also affected by the length of the phrases that the knowledge engineer speaks.

This fact was established by American scientists - linguist Ingve and psychologist Miller. It turned out that a person best perceives sentences with a depth (or length) of 7 plus or minus 2 words. This number (7+2) is called the Yngve-Miller number. It can be considered a measure of the “colloquiality” of speech.

No one doubts the need to fix the procedure for extracting knowledge. The question arises: in what form should this be done? There are three ways to record results:

recording on paper directly during the conversation (disadvantages - this often interferes with the conversation, in addition, it is difficult to have time to write down everything, even if you have shorthand skills);

tape recording, which helps the analyst analyze the entire course of the session and his mistakes (disadvantage - it can hamper the expert);

memorization followed by recording after the conversation (disadvantage - only suitable for analysts with a brilliant memory).

Cognitive layer

Cognitive psychology (eng. cognition) studies the mechanisms by which a person understands the world around him.

Let us offer some advice to the knowledge engineer from the perspective of cognitive psychology:

do not impose on the expert the model of representation that is more understandable and natural to him (the analyst);

use various methods of working with an expert based on the condition that the method should fit the expert like a “key to a lock”;

clearly understand the purpose of the extraction procedure or its main strategy, which can be defined as identifying the main concepts of the subject area and the relationships connecting them;

more often draw diagrams reflecting the expert’s reasoning. This is due to the figurative representation of information in human memory.

The material presented above is closely related to the basics of psychological culture, which includes understanding and knowledge of oneself and other people; adequate self-esteem and assessment of other people; self-regulation of mental state. It is easier to master this culture with the help of specialists - psychologists, psychotherapists, but you can do it yourself with the help of books, at least popular ones, for example. In addition, mastering the basics of acting and participating in special classes in socio-psychological video training contributes to successfully overcoming psychological failures.

In conclusion, here are a number of traditional psychological failures of a novice analyst:

lack of contact between the expert and the knowledge engineer (due to the psychological characteristics of one or the other; errors in the procedure; the occurrence of a “façade” effect, i.e. the expert’s desire to “show himself off”);

lack of understanding (due to the “projection” effect, i.e. transferring the analyst’s view to the expert’s views; or the “order” effect, i.e. focusing attention primarily on what is expressed first, etc.);

low effectiveness of conversations (weak motivation of the expert, i.e. lack of interest; or poor pace of conversation; or inappropriate form of questions; or unsatisfactory answers from the expert).

LINGUISTIC ASPECT

Structure of the linguistic aspect

Since the process of communication between a knowledge engineer and an expert is linguistic communication, let’s consider linguistic aspect knowledge engineering. Let us highlight three layers of linguistic problems important for engineering knowledge (Fig. 17.5).

Rice. 17.5. Structure of the linguistic aspect of knowledge retrieval

Shared code problem

Most psychologists and linguists believe that language is the main means of thinking, along with other sign systems of “internal use.” The languages spoken and thought by the analyst and the expert may differ significantly.

So, we are interested in two languages - analyst language, consisting of three components:

terms of the subject area, which he learned from specialized literature during the preparation period;

general scientific terminology from his “theoretical baggage”;

everyday spoken language used by the analyst;

and language expert consisting:

from special terminology adopted in the subject area;

general scientific terminology; everyday language;

neologisms created by the expert during his work (his professional jargon).

If we assume that the everyday and general scientific languages of the two participants in communication approximately coincide, then some common language, or code, which the partners need to develop for successful interaction, will consist of the flows presented in Fig. 17.6. Subsequently, this general code is transformed into a certain conceptual (semantic) network, which is a prototype of the knowledge field of the subject area.

Rice. 17.6. Scheme for obtaining a common code

The development of a common code begins with the analyst writing down all the terms used by the expert and clarifying their meaning. In fact, this is the compilation of a dictionary of the subject area. This is followed by the grouping of terms and the selection of synonyms (words that mean the same thing). The development of a common code ends with the compilation of a dictionary of subject area terms with a preliminary grouping of them by meaning, i.e. by conceptual proximity (this is the first step in structuring knowledge).

Rice. 17.7 gives an idea of the ambiguity in the interpretation of terms by two specialists. In semiotics, the science of sign systems, the problem of interpretation is one of the central ones. Interpretation connects the “sign” and the “signified object”. Only through interpretation does a sign gain meaning. So, in Fig. 17.7 the words “device X” for an expert mean some specific circuit that corresponds to the circuit of the original device, but in the head of a novice analyst the words “device X” evoke an empty image or some kind of “black box” with handles.

Rice. 17.7. Ambiguity of the problem of interpretation

Conceptual structure

Most experts in artificial intelligence and cognitive psychology believe that the main feature of natural intelligence and memory in particular is the connection of all concepts into a network. Therefore, to develop a knowledge base, you need not a dictionary, but an encyclopedia in which all terms are explained in dictionary entries with links to other terms.

Thus, the linguistic work of a knowledge engineer at a given problem layer is to construct such related fragments by “stitching” terms. With the careful work of the analyst and expert, a hierarchy of concepts begins to emerge in conceptual structures, which is generally consistent with the results of cognitive psychology.

Hierarchy of concepts is a global scheme that can be the basis for a conceptual analysis of the knowledge structure of any subject area.

It should be emphasized that the work of compiling a dictionary and conceptual structure requires linguistic “feeling”, ease of manipulation of terms and a rich vocabulary of a knowledge engineer, since the analyst is often forced to independently develop a dictionary of features. The richer and more expressive the shared code, the more complete the knowledge base.

The analyst is forced to constantly remember the difficulty of conveying images and ideas in verbal form. Often a knowledge engineer has to suggest words and expressions to an expert.

User Dictionary

Linguistic results, correlated with layers of common code and conceptual structure, are aimed at creating an adequate knowledge base. However, we should not forget that the professional level of the end user may not allow him to use the domain-specific language in full. To develop a user interface, additional refinement of the common code dictionary is necessary, adjusted for the accessibility and “transparency” of the system.

In conclusion, we list the characteristic linguistic failures that await a novice knowledge engineer:

speaking different languages (due to the knowledge engineer’s poor training);

mismatch with context and inadequate interpretation of terms (due to lack of feedback, i.e. too independent work of the knowledge engineer);

lack of differences between the general code and the user’s language (differences in the level of knowledge of the expert and the user are not taken into account).

GNOSEOLOGICAL ASPECT

The essence of the epistemological aspect

Epistemology is a branch of philosophy associated with the theory of knowledge, or the theory of the reflection of reality in human consciousness.

Knowledge engineering as a science, so to speak, is doubly epistemological - reality (O) is first reflected in the consciousness of the expert (M 1), and then the activities and experience of the expert are interpreted by the consciousness of the knowledge engineer (M 2), which already serves as the basis for the construction of the third interpretations (P z) – knowledge fields of the expert system (Fig. 17.8). The process of cognition is essentially aimed at creating an internal representation of the surrounding world in the human mind.

Rice. 17.8. Epistemological aspect of knowledge extraction

In the process of knowledge extraction, the analyst is mainly interested in the knowledge component associated with the non-canonical individual knowledge of experts, since subject areas with this type of knowledge are considered most susceptible to the implementation of expert systems. These areas are usually called empirical, since they have accumulated a large volume of individual empirical facts and observations, while their theoretical generalization is a matter for the future.

Knowledge is always associated with the creation of new concepts and theories. It is interesting that often an expert generates new knowledge “on the fly”, right in the context of a conversation with an analyst. Such generation of knowledge can also be useful to the expert himself, who until that moment may not have been aware of a number of relationships and patterns of the subject area. The analyst, who is the “midwife” at the birth of new knowledge, can be helped here by the tools of system methodology, which allows the use of well-known principles of the logic of scientific research and the conceptual hierarchy of science. This methodology forces him to see the general behind the particular, i.e. build chains:

FACT  GENERALIZED FACT  EMPIRICAL LAW  THEORETICAL LAW

The knowledge engineer will not always reach the last link of this chain, but the very desire to move can be extremely fruitful. This approach is fully consistent with the structure of knowledge itself, which has two levels:

empirical (observations, phenomena);

theoretical (laws, abstractions, generalizations).

Criteria of scientific knowledge

Theory is not only a coherent system for generalizing scientific knowledge, it is also a certain way of producing new knowledge. The main methodological criteria for scientific character, which allow us to consider both new knowledge itself and the method of obtaining it scientific, are:

internal consistency and consistency;

consistency;

objectivity;

historicism.

Internal consistency. At first glance, this criterion simply does not work in empirical areas: in them, facts often do not agree with each other, definitions are contradictory, diffuse, etc. An analyst who knows the features of empirical knowledge, its modality, inconsistency and incompleteness, has to smooth out these “roughnesses” of empirics.

Modality of knowledge means the possibility of its existence in various categories, i.e. in the constructions of existence and obligation. Thus, some of the patterns are possible, others are obligatory, etc. In addition, we have to distinguish between such shades of modality as: the expert knows that...; the expert thinks that...; the expert wants...; the expert believes that...

Possible inconsistency empirical knowledge is a natural consequence of the basic laws of dialectics, and these contradictions should not always be resolved in the field of knowledge, but on the contrary, it is the contradictions that most often serve as the starting point in the reasoning of experts.

Incompleteness knowledge is associated with the impossibility of a complete description of the subject area. The analyst’s task is to limit this incompleteness to a certain framework of “completeness”, i.e. narrow the boundaries of the subject area, or introduce a number of restrictions and assumptions that simplify the problem.

Systematicity. The system-structural approach to knowledge (going back to Hegel) orients the analyst to consider any subject area from the standpoint of the laws of the systemic whole and the interaction of its constituent parts. Modern structuralism comes from a multi-level hierarchical organization of any object, i.e. all processes and phenomena can be considered as many smaller subsets (features, details) and, conversely, any objects can (and should) be considered as elements of higher classes of generalizations.

Objectivity. The process of cognition is deeply subjective, i.e. it essentially depends on the characteristics of the knowing subject itself. Subjectivity begins with the description of facts and increases as the idealization of objects deepens.

Consequently, it is more correct to talk about the depth of understanding than about the objectivity of knowledge. Understanding is co-creation, the process of interpreting an object from the point of view of the subject. This is a complex and ambiguous process that takes place in the depths of human consciousness and requires the mobilization of all intellectual and emotional abilities of a person. The analyst should focus all his efforts on understanding the problem. Psychology confirms the fact that people who quickly and successfully solve intellectual problems spend most of their time understanding it, while those who quickly start looking for a solution most often cannot find it.

Historicism. This criterion is related to development. Knowledge of the present is knowledge of the past that gave birth to it. And although most expert systems provide a “horizontal” slice of knowledge - without taking into account time (in statics), a knowledge engineer must always consider processes taking into account time changes - both the connection with the past and the connection with the future. For example, the structure of the knowledge field and the knowledge base must allow adjustment and correction both during the development period and during operation of the expert system.

Structure of cognition

Having considered the main criteria for the scientific nature of knowledge, we will now try to describe its structure. The methodological structure of cognition can be presented as a sequence of stages (Fig. 17.9), which we will consider from the perspective of a knowledge engineer.

Description and synthesis of facts. This is like the “dry residue” of conversations between an analyst and an expert. Carefulness and completeness of keeping records during the extraction process and punctual “homework” on them are the key to a productive first stage of cognition.

In practice, it turns out to be difficult to adhere to the principles of objectivity and consistency described above. Most often, at this stage, facts are simply collected and, as it were, thrown into a “common bag”; An experienced knowledge engineer often immediately tries to find a “shelf” or “box” for each fact, thereby implicitly preparing for the conceptualization stage.

Rice. 17.9. Structure of cognition

Establishing connections and patterns. In the expert's head, connections are established, although often implicitly; The engineer’s task is to identify the framework of the expert’s conclusions. When reconstructing an expert’s reasoning, a knowledge engineer can rely on the two most popular theories of thinking – logical and associative. At the same time, if the logical theory, thanks to its ardent admirers in the person of mathematicians, is widely cited and exploited in every possible way in works on artificial intelligence, then the second, associative theory, is less known and popular, although it also has ancient roots. The beauty and harmony of logical theory should not obscure the sad fact that people rarely think in terms of mathematical logic.

Associative theory represents thinking as a chain of ideas connected by common concepts. The main operations of such thinking are associations acquired on the basis of various connections; recalling past experiences; trial and error with occasional success; habitual (“automatic”) reactions, etc.

Construction of an idealized model. To build a model that reflects the subject’s understanding of the subject area, a specialized language is needed with which one can describe and construct those idealized models of the world that arise in the process of thinking. This language is created gradually with the help of the categorical apparatus adopted in the corresponding subject area, as well as formal symbolic means of mathematics and logic. For empirical subject areas, such a language has not yet been developed, and the field of knowledge, which the analyst will describe in a semi-formalized way, may be the first step towards creating such a language.

Explanation and prediction of models. This final stage of the structure of knowledge is at the same time a partial criterion for the truth of the acquired knowledge. If the identified expert knowledge system is complete and objective, then on its basis it is possible to make predictions and explain any phenomena from a given subject area. Typically, knowledge bases of expert systems suffer from fragmentation and modularity (unrelatedness) of components. All this does not allow us to create truly intelligent systems that, like humans, could predict new patterns and explain cases not explicitly stated in the database. The exception here is knowledge generation systems that are focused on generating new knowledge and “prediction”.

In conclusion, we list the most common failures associated with epistemological problems of knowledge engineering (partially from):

scrappy, fragmented knowledge (due to violations of the principle of consistency or errors in choosing the focus of attention);

inconsistency of knowledge (due to the natural inconsistency of nature and society, incompleteness of the extracted knowledge, incompetence of the expert);

misclassification (due to incorrect determination of the number of classes or inaccurate description of the class);

erroneous level of generalization (due to excessive detail or generalization of object classes).

The system is an intermediary, concluding a supply agreement.

Knowledge engineering is a field of computer science within which research is carried out on the representation of knowledge in computers, keeping it up to date and manipulating it.

Knowledge system - a system based on knowledge.

SOZ SBZ DBMS ES IS SII - artificial intelligence system.

Structure of a knowledge-based system.

KB mechanism for obtaining a solution

INTERFACE

A knowledge base is a model that represents in a computer the knowledge accumulated in a certain subject area. This knowledge must be formalized.
Knowledge is formed using a model and then represented using a specific language.

Knowledge about specific objects and rules are usually highlighted in a knowledge base. These rules are executed as a mechanism for obtaining solutions in order to derive new ones from the original facts.

The interface provides dialogue in a language familiar to the user.

Inference-based methods are often used in knowledge engineering.

The concept of a subject area.

An object is something that exists or is perceived as a separate entity.

Basic properties: discreteness; difference.

When presenting knowledge, a pragmatic approach is used, i.e. those properties of the object that are important for solving the problems that the created system will solve are highlighted. Therefore, a knowledge-based system deals with things that are abstract objects. The object acts as a carrier of some properties of the object. The state of the subject area may change over time. At each moment in time, the state of the subject area is characterized by a set of objects and connections. The state of the subject area is characterized by a situation.

Conceptual means of describing the subject area.

The conceptual model reflects the most general properties. In order to provide a detailed description, languages are needed. The characteristic features of conceptual means of describing a subject area are abstraction and universality. They can be used to describe any subject area.

The concept of an object class.

The concept of an object is the concept of sets. Objects that are similar to each other are combined into classes. At different points in time, different sets of objects can correspond to the same class.

K – object class.

Kt – set of objects of class K at time t.

Group (1999) = (IA-1-99, IA-1-98, …, IA-1-94, IB-1-99,…)

Group (1998) = (IA-1-98, IA-1-97, …, IA-1-93, IB-1-98,…)

(t Кt = ( … )

Teaching position = (professor, associate professor, senior lecturer, lecturer, assistant)

1 4 Geometric figure, square shape, blue color.

(К: А1 К1, А2К2, …, АnКn) name attribute name of classes of object classes attribute pair

Identification of objects can be direct and indirect. In the case of a direct line, the names of objects and serial numbers of objects are used; indirect is based on the use of object properties.

An attribute can be a component. An attribute is understood as a property, characteristic, or name of components.

(Geometric shape: shape Geometric shape color Color)

Attribute name and attribute value pairs are often the same.

Example situation:

(Lecture: lecturer Last name of lecturer, place No. of audience, topic Title of topic, listener Group code, day Day of week, time Start time)

Situation – the connection between “teacher” and “listener” and other characteristics of this situation are shown.

Roles of the participants in the situation:

Listener

Characteristics of the situation:

(K: А1К1, А2К2, …, АnКn) – representation of knowledge in the form of some structure.

(date, day, day_of-month)

(date, month, month_name)

(date, year, year)

(geometric_figure, shape, geometric_shape)

(geometric_figure, color, color)

This representation of knowledge corresponds to the representation of knowledge in the form of individual facts.

(K: A1K1, A2K2, ..., AnKn)

Representations of knowledge about objects are divided into:

object classes (data structure)

knowledge about specific objects (about data)

Object classes.

1. (K: A1K1, A2K2, ..., AnKn)

Аi – attribute name

Ki – object classes, are the attribute value

K – class name

(teachers:

Full name surname_with_initials,

Position teaching_position)

(teacher, full name, surname_with_initials, teacher, teaching_position)

3. K (K1, K2, ..., Kn)

4. K (A1,A2, ..., An)

(teacher (surname_with_initials, teaching_position), teacher (full name, position))

Knowledge representation for the first form:

(K: A1K1,A2K2, ... , AnKn) ki (Ki

Attributive representation of knowledge:

(teacher: - represents

Full name Semenov - some structure

Position assistant professor) - data

Knowledge representation for the second form:

(K: AiKi) k (K, ki (Ki

Attributive representation of knowledge in the form of individual facts:

(teacher1, full name, Semenov) - 1, 2 are links between

(teacher1, position, associate professor) - facts

(teacher2, full name, Petrov)

(teacher2, position, assistant)

Knowledge representation for the third form:

K (K1, K2, … , Kn)

(teacher (Semenov, associate professor) - positional representation of knowledge

If there are no attribute names, and the attributes themselves are written at certain positions, then this is a positional representation of knowledge.

Representation of knowledge in the form of “triples” - (object, attribute, value).

To represent inaccurate values, confidence coefficients are used - (object, attribute, value, confidence coefficient).

0 – corresponds to uncertainty. negative value – the degree of confidence in the impossibility of the attribute value.

(patient 1, diagnosis, gastritis, K740)

* (patient, full name, Antonov, diagnosis colitis K760, gastritis K740)

A representation of knowledge about an object class is called minimal if, when one of the attributes is removed, the remaining set of attributes ceases to be a representation of this object class.

Lease (lease_object, tenant, lessor, lease_term, fee).

If you remove “lease_term”, you get a purchase and sale, and if you remove
“rent_term” and “fee”, then you get a gift.

Representation of knowledge in a relational database.

Relational database – data is stored in a positional format.

The data is stored in the form of a table, where the table name is the name of the class.
Each class corresponds to a table or database file. Class name is the name of the corresponding table. Attribute names – corresponding table fields
(column). Table rows are database records. The entry corresponds to an entry in positional format.
|A1 |A2 | . . .|An |
| | |. . | |
|K1 |K2 | . . .|Кn |
| | |. . | |

Teachers

The concept of an attribute in a positional database is preserved.

The entry K (A1,A2, ..., An) is called a relationship between attributes. This terminology is used in a relational database. The idea of data in a relational database is based on the concept of a “key”.

A key is a set of relation attributes whose value uniquely identifies a record in a file.

Apartment

| city |street |house |building|apartment|area |number of rooms|
|Moscow |Tverskaya |2 |1 |47 |60 |2 |
|Moscow |Tverskaya |2 |1 |54 |50 |1 |

In this case, the key will consist of several fields.

Ki sup Kj is a subclass of class sup subclass; subclass sup class.

Ki is a subclass of Kj if (t Ki t (Kj t

(If at any time t the class Ki is a subclass of Kj)

Npr – network classification.

Network classification is presented as a hierarchical structure.

Student sup student.

Ki part of Kj - is part of Ki part Kj

Ki is part of Kj if a particular object of class Ki is part of a uniquely defined object Kj.

Attitude of belonging. k isa K - is an element

Ki ius K - is a component

Means that an object of class K consists of objects of class K1, K2, ...,
Kn, and an object of class K may include several objects of class Ki.

Lecture No. 4.

Properties of relationships.

Partial order relations have the property of transitivity.

Ki sup Kj Kj sup Km

Ki part Kj Kj part Km

If an element is a component of a block, and the block is composed...

There are no cycles in the membership graph.

K1 ins K2, K2 ins K3,…,Kn-1 ins Kj

It is not true that Kn ins K1

Moscow isa city

City sup Locality

Moscow isa Locality

Operations on classes of objects.

Using operations on object classes, you can define a new object class

Ki set of blocks, for example, TVs

Material objects are divided into three classes

Condition (Premises (Equipment = Material object

Person (Room = Person (Equipment = Room (
Equipment =?

Placing Object Classes

Person (Last name, First name, Patronymic name, Year of birth, gender)

Gender=(male, female)

Man, woman = Human gender

K (K1, K2, K3, K4, K5)

KK5 – Class breakdown by K5 class.

The union of all these classes is man.

Man?Woman=Human

Man?Woman=?

(Knowledge of foreign language

Knowledgeable person

Subject foreign_language)

As a result of the division, we obtain classes of people who know a foreign language.

The conceptual diagram of a subject area is a set of classes of objects, relationships and operations defined on it.

Template descriptions of the state of the subject area:

Classes K conducts discipline classes in a group in on in.

Ivanov I.I. conducts classes in the TOE discipline in group IT-1-98 on Monday on the 4th pair in G-301.

(classes: teacher Teacher discipline Discipline_name group Group_code day Week_day time Pair_number place Audience)

Conceptual models of a subject area - a conceptual diagram along with a set of statements built according to a finite set of templates.

Entity and Relationship Diagram (ER Diagram)

Entity Relation Diagram

Essence

Entity and Relationship Attributes

N teachers work at 1 department. “*” is the sign of the teacher - you can find the department.

Communication verb or object

Attributes – adjective, numerators, dimensions, place of action

Load schedule

Logical systems (models), based on a single example of delivering goods to a store.

Logical models of knowledge representation.

Description of the subject area in one of the logical programming languages, based on predicate calculus.

Language of multiple predicate calculus of the 1st order. Multiple 1st order logic.

To compose this language:

The concept of a sort corresponds to the concept of classes of objects.

Many varieties of S

On the set are specified by functions. f-function name;

types of arguments;
B – type of function value.
Z – signature is the top level of knowledge representation in logical models.

Predicate -
T=(0;1)

false true
- constant of sort B

Let's look at the processing of parts in production as examples.
2-turn;
1-milling;

S=(Part, Machine, Operation, Part_type, Machine_type, Time)
1) child: Operation Detail; f A1 B
2) st: Operation (Machine;
3) start: Operation (Time
4) con: Operation (Time
5) part_type: Part (Part_type
6) type_st: Machine (Type of machine
7) 0: (Time

T: (Time
8) shaft_shaft: (Part_type shaft_place: (Part_type
9) milling cutters: (Machine_type current: (Machine_type
10) cutter_face: operation T current_rev: operation T
11) +: Time*Time Time
12): Time*Time T

Knowledge about specific objects
(lower level of knowledge representation) in the language of multiple predicate calculus is called a structure integrated signature
1) signature
2) Structure of integration. Signatures.
3) For each variety name, many objects of this variety are created.
Part = (part 1, part 2, part 3, part 4)
Machine = (st.1, st.2, st.3)
Operation =(oper1,oper2,oper3,oper4,oper5,oper6,oper7,oper8)
Part_type = (stem_shaft, shaft_place)
Machine_type = (current, milling cutters)
Time = (1,2,…,t)

The union of all sets is the universe.
Each function and predicate from the structure in the system corresponds to many factors.
1) child(oper.1)=child1 child(operation2)=child1 child(operation3)=child2

…………………..
2) st.(oper.1)= st.3 st.(oper.2)= st.1 st.(oper.3)= st.3

…………………
3) start(oper.1)=0 start(operation2)=5 start(operation3)=5
…………………..
4) conc(oper.1)=5 conc(oper.2)=12 conc(oper.3)=0
…………………
5) type_det(part.1)=stem_shaft type_detail(detail.2)=shaft_places type_detail(details3)=steel_shaft type_detail(details4)=shaft_seats
………………….
6) type_st. (st.1)=current type_st. (st.2)=current type_st. (st.3) = cutters
………………….
10) cutter_face (oper1) current_rev (oper2) cutter_face (oper3)
|operation|part |machine |beginning |end |mill_end|current_arr|
|Oper1 |Det.1 |St.3 |0 |5 |1 |0 |
|Oper2 |Det.1 |St.1 |5 |12 |0 |1 |
|Oper3 |Det.2 |Art.3 |5 |10 |1 |0 |
|Oper4 |Det.2 |Art.2 |10 |17 |0 |1 |
|Oper.5 |Det.3 |Art.3 |10 |16 |1 |0 |
|Oper6 |Det.3 |Art.1 |16 |26 |0 |1 |
|Oper7 |Det.4 |Art.3 |16 |22 |1 |0 |
|Oper8 |Det.4 |Art.2 |22 |32 |0 |1 |

|Machine|Type_st |
|St.1 |Current. |
|Art.2 |Current. |
|St.3 |Fr. |

3) Component: Logical formulas

Rules for constructing formulas: a) a constant of sort A is a term of sort A b) a variable taking a value from sort A is a term of sort A c) if the signature contains a function - constructed terms of sorts respectively, then
- there is a term of sort B d) if the signature contains a predicate -
,therms of built varieties
, that is, an atom. e) if - terms of the same kind, then the expression, that is, an atom e) An atom is a correctly constructed formula (PPF) A variable included in an atom is free in this atom. g) if the constructed formula freely includes variables x of type A, then the expressions:

Also is a PPF, the variable “x” is bound (in new files) h) if formulas have already been constructed, then , is also a PPF
Examples:
1) Representation Knowledge b => oper2 performed on a lathe type_st(st(oper2))=ncurrent
2) Opera 2 is completed on stop 1 on st. 1 start 5 end 12
3)

Lecture 8 11/12/99.

Resolution method

The resolution method proves impracticability.
To use this method, it is necessary to convert the original formula to DNF.
DNF:
- disjunction of letters pii – atom or negation of an atom.
Then DNF is represented as a set of clauses
In the resolution method there is one rule of inference
As a result, from 2 clauses we get a new one, called ruoventa
- we get an empty clause, which is always false.
If a set contains an empty clause, then it is unsatisfiable.
The result is an empty clause, which proves that this set is unsatisfiable.
The resolution method is applied until an empty clause is obtained.
m,n – const
substitution instead of a constant variable – unification.
In this case, we perform the substitution (n/y):
From (1) and (2) => a(x)c(x,n) (5)
From (3) and (5), performing the substitution (m/n) => c(m,n) (6)
From (4) and (6) without substitutions => 0

The principle of resolutions in Prolog
Prolog uses Chordian clauses, i.e. clauses containing one letter without negation.
For example
=>

conjunction without negation

Disjuncts that do not contain letters at all can be used. – this is the target statement in the prologue: ? – a a: - b,c,d. b: - e,f. c. e. f.
?-a a(1) a(2) a(3)
|Step No. |Target |Initial |resolution|
| | disjunct | disjunct | |
|1 |?- a. |a:-b,c,d. |-b,c,d. |
|2 |?-b,c,d |b:-e,f |-e,f,c,d |
|3 |?-e,f,c,d |e |-f,c,d |
|4 |?-f,c,d |f |-c,d |
|5 |?-c,d |c |-d |
|6 |?-d |d |0 |

Representation of the program in the form of a graph a: - b;c b: - d,e c: - g,f. e: - i,h g: - h,j d. f. h.
?-a
"," - And
";" - or
The construction of the graph begins with the target clause.
The graph shows which and how many solutions the problem under consideration has.

Two solutions to the problem

Production model of knowledge representation.
The basis for this model is production rules, which have the following form
- production rule >:=
If then [CD=]

Examples:
Rule 5
If gender=female

And addition=small

And weight=65 years_or_more
Then relative_weight = variable
The confidence factor is determined by the number 0-100

Rule 27
IF prospect=excellent

And risk=high
TO factor=0 CD=10
In general, the premise can be a logical expression.
If the premise is true, then the conclusion is true, i.e. the conclusion may indicate some action that is performed if the premise is true
::[AI...I]
::== object, attribute, value, confidence factor - representation of knowledge in the form of a four
::==
:==CD=
The same object can have different meanings.
Multi-valued objects are objects that can have several valid values.
If an object is not declared as multi-valued, then it can have several values, then they do not have to be reliable, i.e. CD= 100

For objects, the value that is requested from the user.
What addition?
1. Small
2. Average allowed values
3. Large

What is the age
1. less than 25
2. from 25 to 55
3. more than 55
Confidence factor of the parcel=min(Kdusl)

Fact obtained as a result of fulfilling the rule prospect = excellent AC = 50 risk = high AC = 70 factor = 0

Basic structure of the production model of knowledge representation

Initial data

Result

Inferences end when the target peak is reached, or there are no applicable rules left and the goal is not achieved.

Reverse conclusions - performed from top to bottom (with conclusions oriented towards the target)

P 1 P2 P3 P4
P5

C 4 C5 C6 C7 C8

|№ |Goal|Conflict |Fulfillment|Subgoals|Fact|
|step| |set | | | |
| | |rules |rules | | |
|1 |C1 |P6,P7 |P6 |S2,C3 | |
|2 |S2 |P1,P2 |P1 |S1,S5,S| |
|3 |C3 | | |3 |F1 |
|4 |C4 | | | |F2 |
|5 |C5 |P3 |P3 | | |
|6 |C6 | | |C6,C7,C|F3 |
|7 |C7 | | |8 |F4 |
|8 |С8 | | | |F5 |
| | | | | | |

Goal – “duration” - the goal is specified by the name of the object.
It is compared with the conclusion of the rules and the rule with the conclusion is selected
, which contain the name of the object. We select a rule that contains the target object, we form a hypothesis

In the process, the hypothesis is either confirmed or refuted. Conclusions continue until one is either confirmed or all possible hypotheses have been exhausted.
Fewer checks are used because a rule has several conditions and one conclusion.

Bidirectional outputs.

First, direct conclusions are made based on a small amount of data, as a result a hypothesis is formed to confirm or refute other conclusions are made.
To check the conditions of the rules, a rule activation apparatus is used, which at each step selects those rules in which the conditions are checked.
Conditions must also be used. In terms of rules, individual rules are distinguished and then general ones.
General rules – rules of conditions of applicability. Scope of applicability.

Generalized structure of a production rule.
(i); Q; P; A; =B; N
(i)– rule name:
Q – scope of application of the rule;
P – condition for applicability of the rule (logical condition)
A=>B – core rule, where A is the premise and B is the conclusion;
N – the set condition determines the actions that are performed if the kernel is executed.
P – if true, the core of the rule is activated.

Frame – a data structure for representing a stereotypical situation
(k: A1K1, A2K2, …., AnKn)
(to: A1k1, A2k2,….,An kn)
(file name: slot1 name (slot1 value) slot2 name (slot2 value)

……………………………….. slot n name (slot n value))
Protoframe – knowledge about a class of objects.
A frame—an instance—is obtained from a protoframe by filling the slots with specific values.
The frame structure usually includes system slots. The slot system includes:
We define slots as a frameparent, a slot pointing to the direct children of a frame.

As a system of slots: slots containing information about the creator of the program and its modification.
The structure includes:
- inheritance indicator;
- data type indicator;
- demons, etc.

FMS LANGUAGE (FMS).
Inheritance pointers can be:
U – unique – unique
S – same- some
R – range – boundary indicator;
0 –override – ignore

U – in frames of different levels with the same names will be different.
S – slots for inheriting values from higher-level slots with the same names

The value of the lower equation must lie within the bounds of the value defined in the upper equation.

R
Human

If the value is not specified, then it is inherited from the top equation slot, and if it is specified, then the inheritance is ignored.

Lecture 11 12/3/99

Combination of network and frame models in the OPS-5 knowledge representation system
This language has production rules and databases
::=({| }+)

()+ - May be repeated several times
::=((value))
::= |
(Substance class acid

Name

Colorless)
(Order of tasks: Source, Leakage Fencing)
What are the rules:
::=(P)
::={}+
::= | -
::= | |
::=((value>)+) |

# (Task order)

([{ }+])
# (Substance)
The pattern does not necessarily indicate all the attributes of a given class, i.e. we can write down
(Substance class acid

Name) i.e. variable acid – the thing will get a value
::= ({ >}+)
The value from the corresponding attribute of the memory work element must match one of the elements specified in this sheet, at least one.
These meanings are specified in specific words.
# (Substance class acid

Color)
::= ({{{}+}}+)
The list of values can also be specified as restrictions
# (Engine power (100 200))

(Engine power 160)
:={}+
::=(make | remove | (modif
{} +)

# (P coordinate _a

(target state active

Name coordinate)
If the target is able to coordinate and the order of tasks is not defined, then create

(Order of tasks) –>
(make target state active

Name order tasks)
(modif1 wait state))

The strategy for solving problems is based on an explicit goal setting
Performance
1. comparison with memory elements resulting in a conflicting set of rules
2. Selection of rules from the conflict set
3. Carrying out the actions specified in the conclusion of the rules
Executes until the goal is achieved.

Acquisition of knowledge

Extracting knowledge from a source, transforming it into the required form, and transferring it to the knowledge base of an intelligent system.

Knowledge is divided into:
- objectified;
- subjective
Objectified - knowledge presented in external sources - books, magazines, research work.
- formatted, i.e. presented in the form of laws, formulas, models, algorithms.
Subjective - knowledge that is expert and empirical is not presented in external form.
An expert’s knowledge is informal, it consists of many heuristic techniques and rules, it allows one to find approaches to solving problems and put forward hypotheses that can be confirmed or refuted.
Knowledge can be obtained in the process of observing any object.
Modes of operation of a knowledge engineer and consultant in the process of acquiring knowledge.
1. protocol analysis
- reasonings are recorded out loud in the process of solving problems.
O.S. protocols are drawn up and analyzed
2. Interview - a dialogue is conducted with the experiment, aimed at acquiring knowledge.
3. Game simulation of professional activity.

Interviewing methods.
1. Chopping into steps identifies connections that allow the construction of hierarchical structures
2. Repertory grid 3 concepts are proposed and it is required to name the difference between the 2nd concept and the 3rd. The expert is offered a couple of concepts and is required to name common properties => form classes.

The method of work of a conitologist to form a field of knowledge
Includes 2 stages
1. preparatory
1.1. Clear preparation of the problem that the system must solve
2. Introducing Konit to Litova
3. Selection of experts
4. Introducing experts to the copy
5. Introducing the expert to a popular artificial intelligence technique
6. Formation of a knowledge field from a copy
2. Main stage
1. pumping the knowledge field in the mode
2. team work of the cosmetologist - analysis of the protocol, identification of connections between concepts, preparation of questions for the expert
3. Pumping up the field of knowledge - the task of asking questions to an expert
4. Formalization of the conceptual problem.
5. Checking the completeness of the model
If the model is incomplete, then the 2nd approximation is used.

Lecture 12 10.12. 99.

Fuzzy sets
– product thickness small medium large

degree

10 15 40 product thickness
- fuzzy set x - universal set
x - form a set of pairs A
- called the membership function of a fuzzy set.
The membership function values for a specific element X are called

Degree of affiliation

Carrier of a fuzzy set
A normal fuzzy set is a set for which

Fuzzy set
X - universal set
X - form a set of pairs A
: - called the membership function of a fuzzy set.
The value of the membership function for a particular element X is called the degree of membership
- carrier of a fuzzy set
&
A normal fuzzy set is a set for each

If we reduce to normal form => we need to divide all its values by
.

Let the membership function be given by an integer from 10 to 40
Define the concept of small thickness of a product.

| | | | | | | | x x

10 11 12 13 14 15 16 17 18
18

Operations on fuzzy sets

1. Union of fuzzy sets

2. Intersection of fuzzy sets

3. Complementation of a fuzzy set

Beginning of lectures 12 and 13.

(A1,(A2,….,(An x1,x2,…,xn x1(X1 x2(X2 … xn(Xn

(A1 x(A2 x … x(An = ()

(x (x1,x2,…,xn) = min((A1 (x1), (A2 (x2)...(An (xn) )

(A x(B = (,
, }
5. Raising a fuzzy set to a power.

(A2 = con((A) - concentration

(A0.5 = dil((A) – stretching

Methods for determining the membership function.

A little more than 2. From 0 to 5.
|x |0 |1 |2 |3 |4 |5 |
|n1 |- |- |- |10|8 |4 |
|n2 |10|10|10|- |2 |6 |

(A = n1 / (n1 + n2)

Ranking method.

Fuzzy variable.

(- name of fuzzy variable x – area of its definition

(A is the meaning, the fuzzy set determines the semantics of the fuzzy variable

Linguistic variable.

(- name of linguistic variable

T – basic term set – forms the names of fuzzy variables
(rarely, sometimes, often), which are linguistic variables

X – carrier of linguistic meanings – domain of definition

G – syntactic procedure

M – semantic procedure

Syntactic procedure in the form of grammatical terms, the symbols of which constitute terms from set terms (and, or, not), type modifiers
(very, slightly, not, etc.)

(- frequency

T = (rarely, sometimes, often)

Often

Such terms, together with the original ones, form a derivative of the terms of the set.

Semantic procedures make it possible to rewrite thermo-fuzzy semantics.

M((1 or (2) = (A1 ((A2

((1, x1, (A1)

((2, x2, (A2)

M((1 and (2) = (A1 ((A2

M(very() = con((A)

M(slightly() = dil((A)

Scenario.

It is a class of frame models for knowledge representation, where knowledge about the sequence of actions and events typical for the subject area is presented in a generalized and structural form. Let's consider a stereotype - a causal scenario - it determines the sequence of actions necessary to achieve goals; this is a frame model.

(kcus name: slot1 name(slot1 value); slot2name(slot2 value);

... slot name n(slot value n))

(kcus actor target actor premise key consequence system name)

The premise specifies the actions that must be performed before the key action in order for it to take effect. Consequence is the final action. System name is script.

(kcus “firefighting”: actor (S:) goal of the actor (C: “stopping the fire”)

P11, P12 parcels (cus: “search for extinguishing agents” R1, “extinguishing vehicles”)

K1 key (f: “use of extinguishing agents for a complete ceasefire”) consequence (P: “ceasefire”) system name (sys: cus*1))

R1 – be earlier

(kcus “search for extinguishing means”: actor (S:) goal of the actor (C: “finding extinguishing means”)

P121, P22 parcels (cus: “determining the coordinates of the location of extinguishing means” R1, “moving to the location of the extinguishing means”)

K2 key (f: “grabbing the extinguishing agents”) consequence (P: “being at the location of the extinguishing agents”) system name (sys: cus*2))

(kcus “transportation of extinguishing means to the place of fire”: actor (S:) goal of the actor (C: “delivery of extinguishing means to the place of fire”)

P31, P32 parcels (cus: “availability of extinguishing means” R1, “determining the coordinates of the fire location”)

K3 key (f: “movement to the fire site”) consequence (P: “finding extinguishing agents at the fire site”) system name (sys: cus*3))

Scenario-based knowledge enhancement.

Sequencing:

D = cus: P11 R1 cus: P12 R1 K1 =

P21R1P22R1K2 P31R1P32R1K3

P21R1P22R1K2 R1 P31R1P32R1K3 R1 K1

Premises define actions that must be performed before the key action and are necessary for its action. The investigation is the final action. System name script.

Replenishment of knowledge based on pseudophysical logics.

P1 – plane landing

P2 – ladder supply

P3 – passengers leaving the plane

P4 – bus delivery

P5 – arrival at the airport terminal

The structure of the text at the linguistic level is represented by the following formula:

TS = PR4dt&P1R3 10,(P2&P2R1P3&P4R3 2,(P5 t = 15 hours 20 minutes

PR4dt , P1R3 10,(P2 (P2R4 dt + 10

P1R3 10,(P2 (P1R1P2

P4R3 2,(P5 (P4R1P5

TS* = P1R1P2& P1R1P3& P2R1P3& P4R1P5

Models and methods of knowledge generalization.

Generalization refers to the process of obtaining knowledge that explains existing facts, as well as the ability to classify, explain and predict new facts. The initial data is represented by a training sample. Objects can be divided into classes. Depending on whether a priori divisions of objects into classes are specified or not, generalization models are divided into generalization models by samples and by classes.

(+ = (01+, 02+…0nj+) – positive sample.

A negative sample can be set (- = (01-, 02-…0ьj-)

It is required to find a rule that allows you to determine whether an object belongs to the class Kj or not.

In data generalization models, a sample is represented by a set of class objects. Generalization methods are divided into generalization methods based on characteristics and structural-logical generalization methods.

Z = (z1, z2, …, zr)

Zi = (zi1, zi2, …, zini)

An object is characterized by a set of feature values Qi = (z1j1, z2j2, …, zrjr).

Structural-logical generalization methods are used to represent knowledge about objects that have an internal structure among structural-logical methods. Two directions can be put forward: inductive methods of normal calculus and generalization methods on semantic networks.

Algorithm for generalizing concepts based on features.

The rules for determining whether objects belong to a certain class are represented in a number of logical formulas whose elements are hij and functions ((((((((

Z = (z1, z2) (gender, age)

Z1 = (z11, z12) (m, f)

Z2 = (z21, z22, z23) (young, middle, old)

(j+ = (01+, 02+) (j- = (01-, 02-, 03-)

01+ = (z11, z21) 02+ = (z11, z22)

01- = (z11, z23) 02- = (z12, z21) 03- = (z12, z22)

&i hij - generalized conjunctive concept

0 = max(xij – 1/(i), where 0 is a criterion, xij is the frequency of occurrence of a certain value of a feature, (i is the number of features.

0 = 3/5 – 1/2 = 0.1

(j+ = (01+, 02+) (j- = (01-)

(-1+ = 0 (-1- = {02-, 03-}

-----------------------

Situation

Static

Dynamic

Constant properties and relationships

States

Sustainable

Temporary

Processes

(patient 1, diagnosis, colitis, K760)
(patient 1, diagnosis, gastritis, K740)

student

Undefined object

Specific object

Material object

Intangible object

situation

space

room

equipment

students

administrative staff

teachers

service staff

graduate student

head of department

Methodist

Professor

assistant

laboratory assistant

space

province

locality

works

teacher

Cafe name

substitution

discipline

Group code

Knowledge engineering is a technology for building expert systems. This process requires a special form of interaction between the creator of the expert system, called a knowledge engineer, and one or more experts in a certain subject area. A knowledge engineer “extracts” from experts the procedures, strategies, and rules of thumb they use to solve problems and embeds this knowledge into an expert system. One of the most difficult problems that arise when creating expert systems is the transformation of the expert’s knowledge and descriptions of the methods he uses to find solutions into a form that allows them to be presented in the system’s knowledge base, and then effectively used to solve problems in a given subject area.

Typically, the expert does not use procedural or quantitative methods. Its main means are analogy, intuition and abstraction. Often the expert cannot even explain how exactly he found the solution. Building a knowledge base includes three stages:
- description of the subject area;
- choice of knowledge representation model;
- acquisition of knowledge.

The first step in building a knowledge base is to identify the subject area for which the expert system is focused on solving problems. In essence, this work comes down to delineating the engineer’s knowledge of the boundaries of the scope of application of the system and the class of problems it solves. In this case it is necessary:
- determine the nature of the tasks to be solved;
- select objects of the subject area;
- establish connections between objects;
- select a knowledge representation model;
- identify specific features of the subject area.

Domain identification is the first step in abstracting the real world. Once a subject area has been identified, the knowledge engineer must formally describe it. To do this, he needs to choose a knowledge representation model. Formally, this should be a model with which you can best display the specifics of the subject area.

The knowledge engineer is first of all obliged to interview the expert and only then begin to build the system. In this case, it is necessary to determine the intended purpose of the system. In this case, the main goal is divided into subgoals.

At the next stage, it is necessary to outline the boundaries of the source data. To build a search space for a solution, it is necessary to define subgoals at each level of the hierarchy of goals of the general problem. At the top of the hierarchy should be placed a task that, in its generality, reflects the fundamental capabilities and purpose of the system.

After identifying the objects of the subject area, it is necessary to establish what connections exist between them. You should strive to identify as many connections as possible.

The resulting qualitative description of the subject area must be represented by means of some formal language in order to bring this description to a form that allows it to be placed in the knowledge base of the system. To solve this problem, a suitable knowledge representation model is selected, with the help of which information about the subject area can be expressed formally.>

The system is an intermediary, concluding a supply agreement.

Knowledge engineering is a field of computer science within which research is carried out on the representation of knowledge in computers, keeping it up to date and manipulating it.

Knowledge system - a system based on knowledge.

SOZ SBZ DBMS ES IS SII - artificial intelligence system.

Structure of a knowledge-based system.

INTERFACE

A knowledge base is a model that represents in a computer the knowledge accumulated in a certain subject area. This knowledge must be formalized. Knowledge is formed using a model and then represented using a specific language.

The interface provides dialogue in a language familiar to the user.

Inference-based methods are often used in knowledge engineering.

The concept of a subject area.

An object is something that exists or is perceived as a separate entity.

Basic properties: discreteness; difference.

Conceptual means of describing the subject area.

The concept of an object class.

K – object class.

K t – set of objects of class K at time t.

Group (1999) = (IA-1-99, IA-1-98, …, IA-1-94, IB-1-99,…)

Group (1998) = (IA-1-98, IA-1-97, …, IA-1-93, IB-1-98,…)

" t K t = ( … )

Teaching position = (professor, associate professor, senior lecturer, lecturer, assistant)

1 4 Geometric figure, square shape, blue color.
objects attribute pair

Identification of objects can be direct and indirect. In the case of a direct line, the names of objects and serial numbers of objects are used; indirect is based on the use of object properties.

An attribute can be a component. An attribute is understood as a property, characteristic, or name of components.

(Geometric figure:

shape Geometric shape

color Color)

Attribute name and attribute value pairs are often the same.

Example situation:

lecturer Last name of lecturer,

place #_audience,

topic Topic_name,

listener Group_code,

day Day of the week,

time Start_time)

Situation – the connection between “teacher” and “listener” and other characteristics of this situation are shown.

Roles of the participants in the situation:

Listener

Characteristics of the situation:

(K: A 1 K 1, A 2 K 2, ..., A n K n) – representation of knowledge in the form of some structure.

(date, day, day_of-month)

(date, month, month_name)

(date, year, year)

(geometric_figure, shape, geometric_shape)

(geometric_figure, color, color)

This representation of knowledge corresponds to the representation of knowledge in the form of individual facts.

(K: A 1 K 1, A 2 K 2, ..., A n K n)

Representations of knowledge about objects are divided into:

Object classes (data structure)

Knowledge about specific objects (about data)

Object classes.

1. (K: A 1 K 1, A 2 K 2, ..., A n K n)

And i is the attribute name

To i – object classes, are the attribute value

K – class name

(teachers:

Full name surname_with_initials,

Position teaching_position)

2. (K: A i K i)

(teacher, full name, surname_with_initials,

teacher, position teaching_position)

3. K (K 1, K 2, ..., K n)

4. K (A 1, A 2, ..., A n)

(teacher (surname_with_initials, teaching_position),

teacher (name, position))

Knowledge representation for the first form:

(K: A 1 K 1 ,A 2 K 2 , … , A n K n) k i Î K i

Attributive representation of knowledge:

(teacher: - represents

Full name Semenov - some structure

Position assistant professor) - data

Knowledge representation for the second form:

(K: A i K i) k Î K, k i Î K i

Attributive representation of knowledge in the form of individual facts:

(teacher 1 , full name, Semenov) - 1 , 2 are links between

(teacher 1 , position, associate professor) - facts

(teacher 2 , full name, Petrov)

(teacher 2 , position, assistant)

Knowledge representation for the third form:

K (K 1, K 2, ..., K n)

(teacher (Semenov, associate professor) - positional representation of knowledge

If there are no attribute names, and the attributes themselves are written at certain positions, then this is a positional representation of knowledge.

Representation of knowledge in the form of “triples” - (object, attribute, value).

To represent inaccurate values, confidence coefficients are used - (object, attribute, value, confidence coefficient).

(patient 1, diagnosis, colitis, K760)

0 – corresponds to uncertainty.

negative value – the degree of confidence in the impossibility of the attribute value.