Intelligent information systems (2) - Abstract. Cheat Sheet: Intelligent Information Systems

Lecture No. 1 IIS

By virtue of their purpose, intelligent information systems can be used in almost any area of ​​human activity. Examples of areas where this approach is already producing tangible results are:

· Industry:

Production management: drawing up and optimizing the production chain by distributing technological steps both between internal departments and between third-party contractors.

Control of production processes: collection and analysis of current information, communication with agents that control other subsystems, adoption and implementation of operational decisions.

Air transport management: modeling and optimization of airport dispatching activities.

Entrepreneurship:

Information management: search for sources, collection, filtering and analysis of data, intellectual processing of large amounts of information.

E-commerce opens up great opportunities for the use of intelligent agents both on the seller's side and on the buyer's side.

Business process management: flexible automation of corporate organizational activities with complex internal logic and a large number of parties involved.

· The medicine:

Patient monitoring: continuous collection, recording and analysis of a large number of monitored characteristics of the patient's condition over an extended period of time.

Healthcare: the possibility of examining and diagnosing patients using virtual specialists from various fields of medicine.

· Entertainment industry:

Computer games: the possibility of reaching qualitatively new levels through the use of intelligent agents for various parties involved.

Interactive applications (television, theater, cinema): agents can create the illusion of the reality of an ongoing action, allowing the user to take part in it.

Examples of IIS in the economy:

· Intelligent Hedger: knowledge-based approach to risk insurance problems. Firm: Information System Department, New York University. The problem of a huge number of ever-growing risk insurance alternatives, rapid decision making by risk managers in an accelerating flow of information, and a lack of appropriate machine support in the early stages of the risk insurance system development process suggests a wide range of different optimal solutions for risk managers. In this system, the development of risk insurance is formulated as a multipurpose optimization problem. This optimization problem involves several difficulties that existing technical solutions cannot cope with. Brief characteristics: the system uses an object representation that captures deep knowledge of risk management and facilitates the emulation of primary risk management reasoning useful for conclusions and their explanations.

· Reasoning system in currency exchange forecasting. Firm. Department of Computer Science City Polytechnic University of Hong Kong Introduces a novel approach to currency exchange forecasting based on accumulation and reasoning with feature support present to focus on a set of hypotheses about exchange rate movements. Represented in the predictive system feature set is a given set of economic values ​​and various sets of time-varying parameters used in the forecasting model. Brief characteristics: the mathematical basis of the applied approach is based on the Dempster-Schaefer theory.

· Nereid. Decision support system for optimizing work with currency options. Firm: NTT Data, The Tokai Bank, Science University of Tokyo. The system facilitates dealer support for optimal response as one of the options presented; more practical and gives better decisions than conventional decision making systems. Brief Features: The system is designed using the CLP frame system, which easily integrates the financial area into the AI ​​application. A mixed type of optimization is proposed, combining heuristic knowledge with linear programming techniques. The system works on Sun-stations.

· PMIDSS: Portfolio Management Decision Support System. Developers: New York University Financial Group. Tasks to be solved: selection of a portfolio of securities; long-term investment planning. Brief characteristics: a mixed knowledge representation system, the use of various inference mechanisms: logic, directed semantic networks, frames, rules.

Currently, the most significant share of the use of intelligent information systems is accounted for by intelligent information agents.

Purpose and functions of intelligent information agents

One of the fundamental concepts in many areas of the theory of artificial intelligence (and, in particular, the planning problem) is the concept of an agent - that object that acts in a certain environment in order to perform certain functions. Despite the widespread use of the term “agent”, there is no generally accepted definition of this concept so far. In the following presentation, the concept of an intelligent agent will be interpreted in the sense of two definitions:

1. Weak definition of an intelligent agent: an intelligent agent is a hardware or software system that has the following properties: autonomy, reactivity, activity and communication.

2. A strong definition of an intelligent agent: an intelligent agent is a computing system that has the listed properties and, in addition, is implemented on the basis of concepts that are most applicable to people.

In this work, the definitions are formulated in a slightly different way in relation to those given above: an agent is understood as an independent software system that has the ability to receive an impact from the outside world, determine its reaction to this impact and carry out this reaction, while the concept of an intelligent agent corresponds to an agent that has a number of knowledge about oneself and the world around and whose behavior is determined by this knowledge

Along with the listed definitions in the literature on artificial intelligence, there are several dozen different formulations of the agent definition, however, most of them come down to the presence of the listed set of key features. Consider the defining properties of intelligent agents in more detail:

Autonomy - the ability to function independently of external control actions (for example, operator control). A high degree of autonomy is facilitated by such agent capabilities as flexible work algorithms, the ability to self-learn, the ability to work with incomplete information.

Reactivity - the ability of an agent to perceive the state of the environment (the outside world) and changes in this state, as well as to take this information into account in its activities. The extreme forms of using the reactivity property are a rigid scheme of the agent's work, in which actions are performed according to a pre-developed plan that is not modified during execution, and a fully reactive behavior scheme, when the agent does not have a pre-prepared plan and acts only on the basis of information about the current state of the environment.

Purposefulness - the ability of an agent not only to reactive actions, but also to purposeful behavior in order to achieve some given goal, set independently or from the outside.

Communicativeness is the property of agents to interact with each other, as well as with other intellectual beings (for example, people). For example, in the problem of distributed artificial intelligence, several agents act in the system, interacting in some way with each other. In a simple case, the interaction is limited only to the exchange of information; in more complex systems, agents can cooperate and adjust their activities to achieve common goals.

An example of an intelligent agent is a softbot (software robot) - a system that interacts with the computer environment (for example, the operating system) by executing commands and interpreting the results of commands and other environment messages.

As practice shows, in most cases, the use of intelligent agents comes down to one of two options:

1. Autonomous performance of specific functions instead of a person, and in some cases, even on behalf of a person.

2. Assistance in the performance of certain activities through high-level human interaction.

As a result of the analysis of known applied systems implemented on the basis of the considered approach, the following types of intelligent agents can be distinguished:

1. Cooperative agents capable not only of autonomous isolated functioning, but also of joint activities with other agents, in particular, coordination of actions, development of common plans and conflict resolution. Examples of agents are Carnegie Mellon University's Pleiades project, MII systems, and ADEPT.

2. Interface agents whose task is to interact with the user (and not with other agents) and help him perform some activity. This type of agent is also sometimes referred to as a personal assistant. Existing implementations include various help systems, trading assistants, workflow support systems, and entertainment systems.

3. Mobile agents that have the ability to perform their functions at different locations within the habitat. The most natural functioning environment for such agents is various computer networks or communication systems. It should also be noted that mobility itself is neither a necessary nor a sufficient property of an intelligent agent.

4. Information agents have emerged as a separate class as a result of a sharp need to search, collect and process a large amount of information with relatively easy access. First of all, this group includes Internet search systems, for example, search in the WWW (Jasper, Webwatcher) and filtering archives of newsgroups (NewT).

5. Reactive agents constitute a special group of agents that do not have any internal model of the environment, but act only in response to a certain state of the environment or a change in state. Examples of such systems are "situational machine", various systems for modeling social behavior, game applications.

6. Hybrid agents that combine the features inherent in various of the above classes. This group includes, in particular, InterRRaP, which combines reactive and cooperative modules, the Guardian patient monitoring system, as well as various mobile information agents.

7. Heterogeneous agent systems, unlike hybrid agents, consist of several agents belonging to different classes. The main motivation for creating such systems is to build the integration of existing specialized systems (ARCHON), while one of the main issues is the organization of interaction between agents.

Despite the obvious advantages and prospects for the use of intelligent agents in various scientific and practical fields, this approach also has a number of limitations, in particular:

· The lack of overall control over intelligent agents can lead to significant difficulties when it comes to global constraints, real-time guaranteed response requirements, and avoiding possible deadlocks.

Lack of a global perspective: since in reality agent systems cannot have a complete system of knowledge about the surrounding world, the question arises of the possibility of finding optimal or suboptimal solutions based on a local knowledge base.

· The problem of trust: how much can you trust intelligent agents when performing their functions autonomously, especially when such an agent acts in the real world on behalf of a certain individual or organization.

However, there is currently an increased demand for smart agent technology from the world's leading corporations, which stimulates a large amount of research work in this area, which is carried out in several main directions:

· The direction on the theory of agents deals with the research and development of specifications, the conceptualization of agents, the definition of properties and ways of their formalized representation. The paper explores the features that distinguish intelligent agents from conventional software, and contains a fairly complete overview of existing theoretical approaches.

· The direction on the architecture of agents studies the implementation of specifications, hardware and software aspects of the problem of building a computing system that satisfies the specified properties. Examples of well-known intelligent agent architectures are: GISA (Generic Intelligent Software Agent) , BDI (Belief-Desire-Intention) and FIPA (Foundation for Intelligent Physical Agents) .

· Direction on programming languages ​​of agents explores ways of formal description of theoretical principles, questions of finding optimal primitives when coding agents, efficient compilation and execution of programs. This group, in particular, includes works devoted to the implementation of agents for the Internet and corporate networks, the study of the features of using the object-oriented approach and the concept of agent-oriented programming.

In accordance with the features listed above, IIS are divided into (this classification is one of the possible) (Fig. 1):

systems with commutative abilities (with intelligent interface);

expert systems (systems for solving complex problems);

self-learning systems (systems capable of self-learning);

adaptive systems (adaptive information systems).

Rice. one. Classification of intelligent information systems by types of systems

Intelligent databases differ from conventional databases in the ability to select, upon request, the necessary information, which may not be explicitly stored, but derived from the database available.

Natural language interface involves the translation of natural language constructions to the intramachine level of knowledge representation. To do this, it is necessary to solve the problems of morphological, syntactic and semantic analysis and synthesis of statements in natural language. So, morphological analysis involves recognizing and checking the spelling of words in dictionaries, syntactic control - decomposing input messages into separate components (defining the structure) with checking for compliance with the grammatical rules of the internal representation of knowledge and identifying missing parts, and, finally, semantic analysis - establishing the semantic correctness of syntactic structures. The synthesis of propositions solves the inverse problem of transforming the internal representation of information into natural language.

The natural language interface is used to:

access to intelligent databases;

contextual search of documentary textual information;

machine translation from foreign languages.

Hypertext systems designed to implement keyword search in text information databases. Intelligent hypertext systems are distinguished by the possibility of a more complex semantic organization of keywords, which reflects the different semantic relationships of terms. Thus, the search mechanism works first of all with the knowledge base of keywords, and only then directly with the text. In a broader sense, the above applies to the search for multimedia information, which includes, in addition to textual, and digital information.

Contextual systems help can be considered as a special case of intelligent hypertext and natural language systems. Unlike conventional help systems that impose on the user a search scheme for the required information, in context help systems, the user describes a problem (situation), and the system, using an additional dialogue, specifies it and searches for recommendations related to the situation itself. Such systems belong to the class of knowledge dissemination systems (Knowledge Publishing) and are created as an annex to documentation systems (for example, technical documentation for the operation of goods).

Cognitive graphics systems allow to carry out the user interface with the IMS using graphic images that are generated in accordance with the events taking place. Such systems are used in monitoring and managing operational processes. Graphic images in a visual and integrated form describe many parameters of the situation under study. For example, the state of a complex controlled object is displayed in the form of a human face, on which each feature is responsible for any parameter, and the general expression of the face gives an integrated description of the situation. Cognitive graphics systems are also widely used in teaching and training systems based on the principles of virtual reality, when graphic images simulate situations in which the student needs to make decisions and perform certain actions.

Expert systems designed to solve problems based on the accumulated knowledge base, reflecting the experience of experts in the problem area under consideration.

Multi-agent systems. Such dynamic systems are characterized by the integration in the knowledge base of several heterogeneous sources of knowledge that exchange the results obtained on a dynamic basis.

For multi-agent systems the following features are characteristic:

carrying out alternative reasoning based on the use of various sources of knowledge with a mechanism for eliminating contradictions;

distributed problem solving, which are divided into parallel solving subproblems corresponding to independent sources of knowledge;

application of a variety of strategies for the operation of the inference mechanism, depending on the type of problem being solved;

processing of large data arrays contained in the database;

use of various mathematical models and external procedures stored in the model database;

the ability to interrupt the solution of problems due to the need to obtain additional data and knowledge from users, models, and subproblems being solved in parallel.

At the core self-learning systems methods of automatic classification of examples of situations of real practice lie.

The characteristic features of self-learning systems are:

self-learning systems “supervised”, when for each example the value of the attribute of its belonging to a certain class of situations (class-forming attribute) is explicitly set;

self-learning systems “without a teacher”, when, according to the degree of proximity of the values ​​of classification features, the system itself identifies classes of situations.

Inductive systems use a generalization of examples according to the principle from particular to general. The process of classifying examples is as follows:

A classification attribute is selected from a set of given ones (either sequentially or according to some rule, for example, in accordance with the maximum number of obtained subsets of examples).

According to the value of the selected feature, the set of examples is divided into subsets.

A check is made whether each resulting subset of examples belongs to one subclass.

If some subset of examples belongs to one subclass, that is, all examples of the subset have the same value of the class-forming attribute, then the classification process ends (while the remaining classification attributes are not considered).

For subsets of examples with a different value of the class-forming feature, the classification process continues from point 1 (each subset of examples becomes a classifiable set).

Neural networks are parallel computing devices consisting of many interacting simple processors. Each processor in such a network deals only with the signals it periodically receives and the signals it periodically sends to other processors.

In expert systems precedent-based(analogies), the knowledge base contains descriptions not of generalized situations, but of the situations themselves or precedents.

Searching for a solution to a problem in precedent-based expert systems is reduced to searching by analogy (that is, abductive inference from particular to particular).

Unlike smart database, information storage is a repository of significant information extracted from the operational database, which is intended for operational situational data analysis (implementation of OLAP technology).

Typical tasks of operational situational analysis are:

determining the profile of consumers of specific storage objects;

prediction of changes in storage objects over time;

analysis of dependences of signs of situations (correlation analysis).

Adaptive Information System is an information system that changes its structure in accordance with the change in the model of the problem area.

Wherein:

an adaptive information system should adequately support the organization of business processes at any given time;

an adaptive information system should adapt whenever there is a need to reorganize business processes;

the reconstruction of the information system should be carried out quickly and at minimal cost.

The core of an adaptive information system is a constantly evolving model of a problem area (enterprise) maintained in a special knowledge base - a repository. Based on the kernel, software is generated or configured. Thus, the design and adaptation of IS is reduced, first of all, to building a model of the problem area and its timely adjustment.

Since there is no generally accepted definition, it is difficult to give a clear unified classification of intelligent information systems. For example, if we consider intelligent information systems from the point of view of the problem being solved, then we can distinguish control systems and reference systems, computer linguistics systems, recognition systems, game systems and systems for creating intelligent information systems (Fig. 2).

At the same time, systems can solve not one, but several problems or, in the process of solving one problem, solve a number of others. For example, when teaching a foreign language, the system can solve the problem of student speech recognition, test, answer questions, translate texts from one language to another, and support a natural language interface.

Figure 2 - Classification of intelligent information systems according to the tasks to be solved

If we classify intelligent information systems according to the criterion "methods used", then they are divided into hard, soft and hybrid (Fig. 3).

Soft Computing is a complex computer methodology based on fuzzy logic, genetic computing, neurocomputing and probabilistic computing. Rigid computing - traditional computer computing (not soft). hybrid systems– systems using more than one computer technology (in the case of intelligent systems, artificial intelligence technologies).

Rice. 3. Classification of intelligent information systems by methods

Other classifications are also possible, for example, general-purpose systems and specialized systems are distinguished (Fig. 4).

Rice. 4. Classification of intelligent systems by purpose

In addition, this diagram reflects another variant of classification by methods: systems using knowledge representation methods, self-organizing systems and systems created with the help of heuristic programming. Also in this classification, music generation systems are classified as communication systems.

To intelligent systems general purpose include systems that not only execute given procedures, but generate and execute procedures for solving new specific problems based on search metaprocedures.

Specialized intelligent systems solve a fixed set of tasks predetermined during system design.

The lack of a clear classification is also explained by the variety of intellectual tasks and intellectual methods, in addition, artificial intelligence is an actively developing science in which new applied areas are mastered daily.

Intelligent information systems

Intelligent Information System(IIS) is one of the types of automated information systems, sometimes IIS is called a knowledge-based system. IIS is a complex of software, linguistic and logical-mathematical tools for the implementation of the main task: the implementation of support for human activities and information retrieval in the advanced dialogue mode in natural language.

IIS classification

• Expert systems
  • Actually expert systems (ES)
  • Interactive banners (web + ES)
• Question-answer systems (in some sources "communication systems")
  • Intelligent search engines (for example, Start system)

IIS can be placed on any site where the user asks the system questions in natural language (if it is a question-answer system) or, by answering the questions of the system, finds the necessary information (if it is an expert system). But, as a rule, ES on the Internet perform advertising and information functions (interactive banners), and serious systems (such as, for example, equipment diagnostics ES) are used locally, as they perform specific specific tasks.
Intelligent search engines differ from virtual interlocutors in that they are rather faceless and, in response to a question, give out some extract from knowledge sources (sometimes quite large), and interlocutors have a “character”, a special manner of communication (they can use slang, profanity), and their answers should be extremely concise (sometimes even just in the form of emoticons, if it fits the context :-)).

For the development of IIS, logical languages ​​were previously used (Prolog, Lisp, etc.), and now various procedural languages ​​are used. Logical and mathematical software is developed both for the modules of the systems themselves, and for joining these modules. However, today there is no universal logical-mathematical system that could satisfy the needs of any IMS developer, so you have to either combine the accumulated experience or develop the system logic on your own. In the field of linguistics, there are also many problems, for example, to ensure the operation of the system in the mode of dialogue with the user in natural language, it is necessary to put algorithms for formalizing natural language into the system, and this task turned out to be much more difficult than expected at the dawn of the development of intelligent systems. Another problem is the constant variability of the language, which must necessarily be reflected in artificial intelligence systems.

Ensuring the work of IIS

• Mathematical
• Linguistic
• informational
• Semantic
• Software
• Technical
• Technological
• personnel

Classification of tasks solved by IIS

• Data interpretation. This is one of the traditional tasks for expert systems. Interpretation refers to the process of determining the meaning of data, the results of which must be consistent and correct. Usually, a multivariate analysis of the data is provided.
• Diagnostics. Diagnostics refers to the process of relating an object to a certain class of objects and/or detecting a fault in a certain system. A fault is a deviation from the norm. This interpretation makes it possible to consider, from a unified theoretical standpoint, equipment malfunctions in technical systems, diseases of living organisms, and all kinds of natural anomalies. An important specificity here is the need to understand the functional structure (“anatomy”) of the diagnostic system.
• Monitoring. The main task of monitoring is the continuous interpretation of data in real time and signaling that certain parameters go beyond the permissible limits. The main problems are the “skip” of an alarming situation and the inverse problem of a “false” alarm. The complexity of these problems lies in the blurring of the symptoms of anxiety situations and the need to take into account the temporal context.
• Design. Design consists of preparing specifications for the creation of "objects" with predetermined properties. The specification is understood as the entire set of necessary documents - a drawing, an explanatory note, etc. The main problems here are obtaining a clear structural description of knowledge about the object and the “trace” problem. To organize effective design and, to an even greater extent, redesign, it is necessary to form not only the design decisions themselves, but also the motives for their adoption. Thus, in design problems, two main processes are closely connected, performed within the framework of the corresponding ES: the process of deriving a solution and the process of explaining.
• Forecasting. Forecasting allows you to predict the consequences of certain events or phenomena based on the analysis of available data. Predictive systems logically deduce likely consequences from given situations. In a predictive system, a parametric dynamic model is usually used, in which the values ​​of the parameters are “fitted” to a given situation. The consequences derived from this model form the basis for forecasts with probabilistic estimates.
• Planning. Planning is understood as finding action plans related to objects capable of performing certain functions. In such ES, behavioral models of real objects are used in order to logically deduce the consequences of the planned activity.
• Education. Learning refers to the use of a computer to teach some discipline or subject. Training systems diagnose errors in the study of any discipline with the help of a computer and suggest the right solutions. They accumulate knowledge about the hypothetical "student" and his characteristic mistakes, then in the work they are able to diagnose weaknesses in the knowledge of the trainees and find appropriate means to eliminate them. In addition, they plan the act of communicating with the student depending on the success of the student in order to transfer knowledge.
• Control. Management is understood as a function of an organized system that supports a certain mode of activity. Such kind of ES control the behavior of complex systems in accordance with the given specifications.
• Decision Support. Decision support is a set of procedures that provides the decision maker with the necessary information and recommendations to facilitate the decision making process. These ES help specialists to choose and/or form the necessary alternative among the many choices when making responsible decisions.

In the general case, all knowledge-based systems can be divided into systems that solve problems of analysis and systems that solve problems of synthesis. The main difference between analysis problems and synthesis problems is that if in analysis problems the set of solutions can be listed and included in the system, then in synthesis problems the set of solutions is potentially unlimited and is built from solutions of components or sub-problems. The objectives of the analysis are: data interpretation, diagnostics, decision support; synthesis tasks include design, planning, and control. Combined: training, monitoring, forecasting.

see also

Links

• Association for Artificial Intelligence on the Internet alicebot.org

Wikimedia Foundation. 2010 .

See what "Intelligent Information Systems" is in other dictionaries:

- (IIT) (eng. Intellectual information technology, IIT) are information technologies that help a person to accelerate the analysis of the political, economic, social and technical situation, as well as the synthesis of management decisions. At the same time ... ... Wikipedia

intelligent systems- INTELLIGENT SYSTEMS (from lat. intellectus mind, mind) computer systems that implement some features of human intelligence, making it possible to master difficult tasks that a person can solve in real time ... ... Encyclopedia of Epistemology and Philosophy of Science

- (IT, from the English information technology, IT) a wide class of disciplines and areas of activity related to technologies for managing and processing data, including the use of computer technology. Recently, under information ... ... Wikipedia

Institute of Automation and Computer Engineering of the Moscow Power Engineering Institute (Technical University) ... Wikipedia

Moscow State University of Instrument Engineering and Informatics (MGUPI) Founded 1936 Rector ... Wikipedia

Moscow State University of Instrument Engineering and Informatics (MGUPI) Founded 1936 Rector ... Wikipedia

Moscow State University of Instrument Engineering and Informatics (MGUPI) Founded 1936 Rector ... Wikipedia

Decision-making regarding actions or behavior in a given situation of any subjects (humans, robots, complex control systems) is carried out on the basis of information processes. The information process implements the relationship between the object and the subject (Fig. 1.1) and represents the subject's perception of objective reality in the form of data, the processing of this data in accordance with the target setting and the available knowledge about the dependencies of facts into information.

Based on the information received, the knowledge of the subject is updated, a decision is made on a possible change in the state of the object and the target setting of the subject. Thus, the information process is considered in three aspects:

The syntactic aspect is the display of objective reality in any environment and in any language, which is data.

The semantic aspect is the understanding and interpretation of data based on the knowledge of the subject, which reflect dependencies, patterns of interaction between objects.

The pragmatic aspect is the assessment of the usefulness of the acquired new knowledge (increment of knowledge) of the subject in accordance with the target setting for making a decision, that is, obtaining information in the narrow sense.

In a broad sense, information is understood as all three aspects of the reflection of the information process.

Any computer information system (IS) that implements an information process performs the following functions: it receives information requests entered by the user (goals for solving the problem) and the necessary initial data, processes the data entered and stored in the system in accordance with a known algorithm, and generates the required output information. From the point of view of the implementation of the listed functions, IS can be considered as a factory that produces information, in which the order is an information request, the raw material is the initial data, the product is the required information, and the tool (equipment) is the knowledge with which the data is converted into information.

The knowledge of the subjects of the information process can be represented in various forms. In humans, knowledge is presented either in an undocumented (implicit, directly in the head) form, or in a documented (explicit, bookish) form. Moreover, a documented textual form of knowledge representation in the form of textbooks, provisions, instructions, etc. it is not well suited for quickly extracting the necessary knowledge when substantiating specific decisions. Implicit knowledge of experts is generally difficult to access for use in solving problems by other specialists.

Computer information systems, acting as subjects of the information process, are designed to simplify the process of using knowledge in solving decision-making problems. To do this, knowledge must be structured and memorized for subsequent repeated use.

Knowledge has a dual nature: factual and operational:

Factual knowledge represents known information about the objects of reflected reality and accumulates in conventional databases.

Operational knowledge reflects the dependencies and relationships between objects that allow you to interpret data or extract information from them. Operational knowledge is presented either in an algorithmic form or in a declarative form in the form of special structured knowledge bases.

Often factual knowledge is called extensional (detailed), and operational knowledge - intensional (generalized).

The information process with the help of a computer information system is reduced to an adequate combination of operational and factual knowledge and is performed differently in various types of information systems. The easiest way to connect them is within a single application program:

Document without a title

Thus, operational knowledge (algorithm) and factual knowledge (data structure) are inseparable from each other. However, if during the operation of the IS it becomes clear that one of the two components of the program needs to be modified, then it will need to be rewritten. This is explained by the fact that only the IS developer has full knowledge of the problem area, and the program serves as a "non-thinking executor" of the developer's knowledge. The end user, due to the procedural nature and machine orientation of knowledge representation, understands only the external side of the data processing process and cannot influence it in any way.

The consequence of these shortcomings is the poor viability of IS or non-adaptation to changes in information needs. In addition, due to the determinism of the algorithms of the tasks being solved, the IS is not capable of forming the user's knowledge about actions in incompletely defined situations. In systems based on the processing of databases (SDB - Data Base Systems), there is a separation of factual and operational knowledge from each other. The first is organized in the form of a database, the second - in the form of programs. Moreover, the program can be automatically generated at the request of the user (for example, the implementation of SQL or QBE queries). A software tool for accessing data - a database management system (DBMS) acts as an intermediary between the program and the database:

Document without a title

SBD = Program<=>DBMS<=>Database

The concept of program independence from data makes it possible to increase the flexibility of the IS to fulfill arbitrary information requests. However, this flexibility, due to the procedural nature of the representation of operational knowledge, has well-defined boundaries. To formulate an information request, the user must clearly understand the structure of the database and, to a certain extent, the algorithm for solving the problem. Therefore, the user must have a good understanding of the problem area, the logical structure of the database and the algorithm of the program. The conceptual database schema acts mainly only as an intermediate link in the process of mapping the logical data structure to the data structure of the application program.

The general disadvantages of traditional information systems, which include systems of the first two types, are weak adaptability to changes in the subject area and information needs of users, the inability to solve poorly formalized tasks that managers constantly deal with. These shortcomings are eliminated in intelligent information systems (IIS).

An analysis of the program structure shows the possibility of extracting operational knowledge (data transformation rules) from the program into the so-called knowledge base, which in a declarative form stores units of knowledge common to various tasks. At the same time, the control structure acquires the character of a universal mechanism for solving problems (inference mechanism), which links units of knowledge into executable chains (generated algorithms) depending on the specific problem statement (formulated in the request of the goal and initial conditions). Such IS become systems based on knowledge processing (KBS - Knowledge Base (Based) Systems):

Document without a title

The next step in the development of intelligent information systems is the separation into an independent subsystem or repository of metaknowledge, which describes the structure of operational and factual knowledge and reflects the model of the problem area. In such systems, both programs and data structures are generated or assembled from the knowledge units described in the repository each time the problem domain model changes. Let's call IIS, processing metaknowledge, systems based on models (SBM - Model Based Systems):

Document without a title

For intelligent information systems focused on the generation of algorithms for solving problems, the following features are characteristic:

Developed communication skills

Ability to solve complex poorly formalized problems,

The ability to self-learn,

Adaptability.

IIS communication skills characterize the way of interaction (interface) of the end user with the system, in particular, the possibility of formulating an arbitrary request in a dialogue with the IIS in a language as close as possible to natural.

Complex poorly formalized tasks- these are tasks that require the construction of an original solution algorithm depending on the specific situation, which may be characterized by uncertainty and dynamism of the initial data and knowledge.

Ability to self-learn- this is the possibility of automatic extraction of knowledge for solving problems from the accumulated experience of specific situations.

adaptability- the ability to develop the system in accordance with objective changes in the model of the problem area.

In various IIS, the listed signs of intelligence are developed to an unequal degree and it is rare that all four signs are realized simultaneously. Conventionally, each of the signs of intelligence corresponds to its own class of IIS (Fig. 1.2):

Systems with intelligent interface;

Expert systems;

Self-learning systems;

adaptive systems.

All four features of intelligence are implemented to some extent in knowledge management systems.

Intelligent databases differ from conventional databases in the ability to select, upon request, the necessary information, which may not be explicitly stored, but derived from those available in the database. The output of implicit information is carried out by interpreting the following dependencies:

Computational dependencies of attributes, for example, "display a list of goods whose price is higher than the industry average",

Structural relations of objects, for example, "display a list of substitute products for some products",

Logical dependencies of decision-making factors, for example, "display a list of potential buyers of a certain product."

To execute the first type of query, it is first necessary to carry out a statistical calculation of the average industry price for the entire database, and only after that, the actual selection of data.

To execute the second type of query, it is necessary to display the values ​​of the characteristic features of the object, and then search for similar objects using them.

For the third type of request, you first need to determine the list of intermediary sellers selling this product, and then search for buyers associated with them.

In all the above types of queries, it is required to search by a condition, which must be further defined in the course of solving the problem. The intelligent system, without the help of the user, uses the structure of the database to build the access path to the data files. The formulation of the request is carried out in a dialogue with the user, the sequence of steps of which is performed in the most convenient form for the user. A database query can also be formulated using a natural language interface.

Natural language interface involves the translation of natural language constructions to the intramachine level of knowledge representation. To do this, it is necessary to solve the problems of morphological, syntactic and semantic analysis and synthesis of statements in natural language. So, morphological analysis involves recognizing and checking the spelling of words in dictionaries, syntactic control - decomposing input messages into separate components (defining the structure) with checking for compliance with the grammatical rules of the internal representation of knowledge and identifying missing parts, and, finally, semantic analysis - establishing the semantic correctness of syntactic structures. The synthesis of propositions solves the inverse problem of transforming the internal representation of information into natural language.

The natural language interface is used to:

Access to intelligent databases;

Contextual search of documentary textual information;

Machine translation from foreign languages.

Hypertext systems designed to implement keyword search in text information databases. Intelligent hypertext systems are distinguished by the possibility of a more complex semantic organization of keywords, which reflects the different semantic relationships of terms. Thus, the search mechanism works first of all with the knowledge base of keywords, and only then directly with the text. In a broader sense, this also applies to the search for multimedia information, which includes, in addition to textual and digital information, graphic, audio and video images.

Context help systems can be considered as a special case of intelligent hypertext and natural language systems. Unlike conventional help systems that impose on the user a search scheme for the required information, in context help systems, the user describes a problem (situation), and the system, using an additional dialogue, specifies it and searches for recommendations related to the situation itself. Such systems belong to the class of knowledge dissemination systems (Knowledge Publishing) and are created as an annex to documentation systems (for example, technical documentation for the operation of goods).

Cognitive graphics systems allow to carry out the user interface with the IMS using graphic images that are generated in accordance with the events taking place. Such systems are used in monitoring and managing operational processes. Graphic images in a visual and integrated form describe many parameters of the situation under study. For example, the state of a complex controlled object is displayed in the form of a human face, on which each feature is responsible for any parameter, and the general expression of the face gives an integrated description of the situation.

Cognitive graphics systems are also widely used in teaching and training systems based on the use of virtual reality principles, when graphic images simulate situations in which the student needs to make decisions and perform certain actions.

Appointment of expert systems consists in solving problems that are rather difficult for experts on the basis of an accumulated knowledge base that reflects the experience of experts in the problem area under consideration. The advantage of using expert systems lies in the possibility of making decisions in unique situations for which the algorithm is not known in advance and is formed from the initial data in the form of a chain of reasoning (decision rules) from the knowledge base. Moreover, the solution of problems is supposed to be carried out in conditions of incompleteness, unreliability, ambiguity of the initial information and qualitative assessments of processes.

An expert system is a tool that enhances the intellectual abilities of an expert and can perform the following roles:

Consultant for inexperienced or non-professional users;

Assistant in connection with the need for an expert to analyze various decision-making options;

An expert partner on issues related to sources of knowledge from related fields.

Expert systems are used in many areas, among which the segment of applications in business is in the lead (Fig. 1.3).

Expert system architecture includes two main components: a knowledge base (repository of knowledge units) and a software tool for accessing and processing knowledge, consisting of mechanisms for deriving conclusions (solutions), acquiring knowledge, explaining the results obtained and an intelligent interface (Fig. 1.4). Moreover, the central component of the expert system is the knowledge base, which acts in relation to other components as a meaningful subsystem that constitutes the main value. The "know-how" of the knowledge base of a good expert system is estimated at hundreds of thousands of dollars, while the software toolkit is in thousands or tens of thousands of dollars.

Knowledge base is a set of knowledge units that represent a reflection of the objects of the problem area and their relationships, actions on objects and, possibly, the uncertainties with which these actions are carried out, formalized with the help of a certain method of knowledge representation.

As methods of knowledge representation, either rules, or objects (frames), or a combination of them are most often used. So, the rules are constructions:

Document without a title

As a rule, the certainty factors (CF) are either the conditional probabilities of the Bayesian approach (from 0 to 1) or the fuzzy logic confidence coefficients (from 0 to 100). Example rules look like this:

Document without a title

Frames are a collection of attributes that describe properties and relationships with other frames. Unlike database entries, each frame has a unique name. Some of the attributes reflect typified relationships, such as "genus - species" (super-class - sub-class), "whole - part", etc. Instead of specific values ​​of object attributes, default values ​​can be set (the attribute inheritance indicator is set to S), inherent in entire classes of objects, or attached procedures (process). An example of frames is shown in fig. 1.5.

Intelligent interface. Data exchange between the end user and the ES is performed by an intelligent interface program that receives user messages and converts them into a knowledge base representation form and, conversely, translates the internal representation of the processing result into the user's format and outputs a message to the required media. The most important requirement for the organization of the user's dialogue with the ES is naturalness, which does not mean literally formulating the user's needs with natural language sentences, although this is not excluded in some cases. It is important that the sequence of solving the problem is flexible, consistent with the user's ideas and conducted in professional terms.

withdrawal mechanism. This software tool receives a request converted into an internal representation from the intelligent interface, generates a specific algorithm for solving the problem from the knowledge base, executes the algorithm, and the result is provided to the intelligent interface to issue a response to the user's request.

The use of any inference mechanism is based on the process of finding units of knowledge (rules, objects, precedents, etc.) relevant to the solution in accordance with the goal and description of a specific situation (initial data) and linking them, if necessary, into a chain of reasoning, leading to a certain result.

To represent knowledge in the form of rules, this can be a direct (Fig. 1.6) or reverse (Fig. 1.7) chain of reasoning.

Frame (object-oriented) representation of knowledge is characterized by the use of the attribute inheritance mechanism, when attribute values ​​are transferred along the hierarchy from higher classes to lower ones (for example, in Fig. 1.5, the industry code, the industry profitability ratio). Also, when the frame attributes are filled with the necessary data, the attached procedures are launched for execution.

explanation mechanism. In the process or as a result of solving the problem, the user can request an explanation or justification for the solution. To this end, the ES must provide an appropriate explanation mechanism. The explanatory abilities of ES are determined by the ability of the inference mechanism to memorize the way to solve the problem. Then the user's questions "How?" and why?" a decision is received or some data is requested, the system can always issue a chain of reasoning up to the required control point, accompanying the issuance of an explanation with pre-prepared comments. In the absence of a solution to problems, an explanation should be given to the user automatically. It is useful to have the possibility of a hypothetical explanation of the solution of the problem, when the system answers questions about what will happen in this or that case.

However, the user may not always be interested in the full output of the solution, which contains many unnecessary details. In this case, the system should be able to select only key points from the chain, taking into account their importance and the level of knowledge of the user. To do this, the knowledge base needs to maintain a model of user knowledge and intentions. If the user continues to not understand the received answer, then the system should be able to teach the user certain fragments of knowledge in a dialogue based on the supported model of problematic knowledge, i.e. disclose individual concepts and dependencies in more detail, even if these details were not used directly in the output.

The mechanism of knowledge acquisition. The knowledge base reflects the knowledge of experts (specialists) in a given problem area about actions in various situations or processes for solving specific problems. The identification of such knowledge and its subsequent presentation in the knowledge base are carried out by specialists called knowledge engineers. To enter knowledge into the database and their subsequent updating, the ES must have a mechanism for acquiring knowledge. In the simplest case, this is an intelligent editor that allows you to enter units of knowledge into the database and carry out their syntactic and semantic control, for example, for consistency. In more complex cases, the knowledge acquisition mechanism makes it possible to extract knowledge as a result of using special scenarios for interviewing experts, or from input examples of real situations, as in the case of inductive inference, or from texts, or from the experience of the intellectual system itself.

Classes of expert systems. According to the degree of complexity of the tasks to be solved, expert systems can be classified as follows:

According to the method of forming a solution Expert systems are divided into two classes: analytical And synthetic. Analytical systems involve the choice of solutions from a set of known alternatives (determination of the characteristics of objects), and synthetic systems - the generation of unknown solutions (the formation of objects).

According to the method of taking into account the temporal sign expert systems can be static or dynamic. Static systems solve problems with data and knowledge unchanged in the process of solving, dynamic systems allow such changes. Static systems carry out a monotonous uninterrupted solution of the problem from the input of initial data to the final result, dynamic systems provide for the possibility of revising the previously obtained results and data in the process of solving.

By types of data and knowledge used expert systems are classified into systems with deterministic(well-defined) knowledge and uncertain knowledge. The uncertainty of knowledge (data) is understood as their incompleteness (absence), unreliability (measurement inaccuracy), ambiguity (ambiguity of concepts), fuzziness (qualitative assessment instead of quantitative).

By the number of sources of knowledge used expert systems can be built using one or sets sources of knowledge. Sources of knowledge can be alternative (many worlds) or complementary (cooperating).

In accordance with the listed features of the classification, as a rule, the following four main classes of expert systems are distinguished (Fig. 1.8)

Classifying expert systems. Analytical tasks primarily include tasks of recognizing various situations, when the essence of a certain situation is revealed by a set of given features (factors), depending on which a certain sequence of actions is selected. Thus, in accordance with the initial conditions, among the alternative solutions there is one that best satisfies the goal and constraints.

Expert systems that solve problems of situation recognition are called classifying, since they determine the belonging of the analyzed situation to a certain class. As the main method of forming decisions, the method of logical deductive inference from general to particular is used, when a particular conclusion is obtained by substituting the initial data into a certain set of interrelated general statements.

Extending expert systems. A more complex type of analytical problems are those that are solved on the basis of uncertain initial data and applied knowledge. In this case, the expert system should, as it were, determine the missing knowledge, and in the decision space, several possible solutions can be obtained with different probability or confidence in the need for their implementation. Bayesian probabilistic approach, confidence coefficients, fuzzy logic can be used as methods of working with uncertainties. Extending expert systems can use several sources of knowledge to form a solution. In this case, heuristic techniques for selecting knowledge units from their conflict set can be used, for example, based on the use of importance priorities, or the resulting degree of certainty of the result, or the values ​​of preference functions, etc.

The following are typical for analytical problems of classifying and redefining types: problem areas:

Data interpretation- choice of a solution from a fixed set of alternatives based on the input information about the current situation. The main purpose is to determine the essence of the situation under consideration, the choice of hypotheses based on their facts. A typical example is an expert system for analyzing the financial condition of an enterprise.

Diagnostics- Identification of the causes that led to the occurrence of the situation. A preliminary interpretation of the situation is required, followed by verification of additional facts, for example, the identification of factors that reduce production efficiency.

Correction- diagnostics, supplemented by the possibility of assessing and recommending actions to correct deviations from the normal state of the situations under consideration.

Transforming expert systems. Unlike analytical static expert systems, synthesizing dynamic expert systems involve a repetitive transformation of knowledge in the process of solving problems, which is associated with the nature of the result, which cannot be predetermined in advance, as well as with the dynamism of the problem area itself.

Varieties of hypothetical inference are used as methods for solving problems in transforming expert systems:

Generation and testing, when hypotheses are generated from the initial data, and then the formulated hypotheses are tested for confirmation by the incoming facts;

Assumptions and defaults, when knowledge about similar classes of objects is selected from incomplete data, which subsequently dynamically adapt to a specific situation depending on its development;

The use of general patterns (metacontrol) in the case of unknown situations, allowing to generate the missing knowledge.

Multi-agent systems. Such dynamic systems are characterized by distributed problem solving by several software agents, each of which has its own knowledge base and inference mechanism. Software agents, as a rule, carry out the instructions of people, the subjects of solving the problem, and in this sense they are replaced. At the same time, they react to events in the external environment (reactive agents), process situations, make decisions, transfer the results of solving problems to users and to the external environment. The most intelligent (cognitive) agents are able to learn and change the rules of their behavior.

When jointly solving problems by several software agents, multi-agent systems (MAC) are formed, with centralized or decentralized control. In the first case, the MAC must have at least one agent that acts as a coordinator (dispatcher), planning and controlling the implementation of processes. In the second case, all agents are independent in their behavior. The integration of the work of software agents and the corresponding sources of knowledge is carried out on a dynamic basis by exchanging the results obtained between them, for example, through a “bulletin board” (Fig. 1.9).

Multi-agent systems are characterized by the following features:

Carrying out alternative reasoning based on the use of various sources of knowledge with a mechanism for eliminating contradictions;

Distributed problem solving, which are divided into parallel solving subproblems corresponding to independent sources of knowledge;

Application of a variety of strategies for the operation of the inference mechanism, depending on the type of problem being solved;

Processing of large amounts of data contained in the database, and the ability to self-learn, changing the rules of behavior of agents;

Use of various mathematical models and external procedures stored in the model database;

The ability to interrupt the solution of problems due to the need to obtain additional data and knowledge from users, models, and subproblems being solved in parallel.

For synthesizing dynamic expert systems, the following are most applicable. problem areas:

Design- defining the configuration of objects in terms of achieving specified performance criteria and constraints, for example, designing an enterprise budget, investment portfolio, product configuration in e-commerce.

Forecasting- prediction of the consequences of the development of current situations based on mathematical and heuristic modeling, for example, forecasting trends in stock trading.

Planning- selection of a sequence of user actions to achieve the goal, for example, planning supply chains of products (supply chain management).

Dispatching- distribution of work in time, scheduling, for example, scheduling workflows (workflow).

Monitoring- tracking the current situation with possible subsequent correction. To do this, diagnostics, forecasting, and, if necessary, planning and correction of user actions are performed, for example, monitoring the sale of finished products.

Control- monitoring, supplemented by the implementation of actions in automatic systems, for example, decision-making at exchange trading.

According to the publication, which analyzes 12,500 existing expert systems, the distribution of expert systems by problem areas is as follows (Fig. 1.10):

Self-learning systems are based on methods for automatic classification of examples of real-life situations (learning by example). Examples of real situations are accumulated over a certain historical period and make up training set. These examples are described by a variety of classification features. Moreover, the training sample can be:

“with a teacher”, when for each example the value of the attribute of its belonging to a certain class of situations (class-forming attribute) is explicitly set;

“without a teacher”, when, according to the degree of proximity of the values ​​of the classification features, the system itself identifies classes of situations.

As a result of system training, generalized rules or functions are automatically constructed that determine the belonging of situations to classes that the trained system uses when interpreting new emerging situations. Thus, a knowledge base is automatically formed, which is used in solving problems of classification and forecasting. This knowledge base is periodically automatically adjusted as experience of real situations is accumulated, which reduces the cost of its creation and updating.

Common disadvantages common to all self-learning systems are as follows:

Incompleteness and/or noise (redundancy) of the training sample is possible and, as a result, the relative adequacy of the knowledge base to emerging problems;

There are problems associated with poor semantic clarity of feature dependencies and, as a result, the inability to explain the results to users;

Limitations in the dimension of the attribute space cause a shallow description of the problem area and a narrow focus of application.

inductive systems. The generalization of examples according to the principle from particular to general is reduced to identifying subsets of examples belonging to the same subclasses and determining significant features for them.

The process of classifying examples is as follows:

1. A classification feature is selected from a set of given ones (either sequentially or according to some rule, for example, in accordance with the maximum number of obtained subsets of examples);

2. According to the value of the selected feature, the set of examples is divided into subsets;

3. A check is made whether each resulting subset of examples belongs to one subclass;

4. If some subset of examples belongs to the same subclass, i.e. all examples of the subset have the same value of the class-forming attribute, then the classification process ends (while the remaining classification attributes are not considered);

5. For subsets of examples with a different value of the class-forming feature, the classification process continues from point 1. (Each subset of examples becomes a classifiable set).

The classification process can be represented as a decision tree, in which the intermediate nodes contain the values ​​of the attributes of a sequential classification, and the final nodes contain the values ​​of the attribute of belonging to a certain class. An example of building a decision tree based on a fragment of the example table (table 1.1) is shown in fig. 1.11.

Table 1.1

Document without a title

class sign	Classification signs
		Competition	Costs	Quality
		small	small
		small
		small
		small	small
		small	small
		small

The analysis of a new situation is reduced to the choice of a tree branch that completely defines this situation. The search for a solution is carried out as a result of a sequential check of the features of the classification. Each branch of the tree corresponds to one decision rule:

If Demand = "low" and Costs = "small"
Then Price = "low"

Examples of tools that support inductive knowledge inference are 1st Class (Programs in Motion), Rulemaster (Radian Corp.), ILIS (ArgusSoft), KAD (IPS Pereyaslavl-Zalessky).

Neural networks. As a result of training on examples, mathematical decision functions (transfer functions or activation functions) are built that determine the dependencies between input (Xi) and output (Yj) features (signals) (Fig. 1.12).

Each such function, called by analogy with the elementary unit of the human brain - a neuron, displays the dependence of the value of the output feature (Y) on the weighted sum (U) of the values ​​of the input features (Xi), in which the weight of the input feature (Wi) shows the degree of influence of the input feature on day off:

selection">Fig. 1.13).

Neurons can be interconnected when the output of one neuron is the input of another. Thus, a neural network is built (Fig. 1.14), in which neurons located at the same level form layers.

Training a neural network comes down to determining the connections (synapses) between neurons and establishing the strength of these connections (weight coefficients). Neural network learning algorithms are simplified to determine the dependence of the weight coefficient of the connection of two neurons on the number of examples that confirm this dependence.

The most common neural network training algorithm is the backpropagation algorithm. The objective function according to this algorithm should ensure the minimization of the squared learning error for all examples:

formula" src="http://hi-edu.ru/e-books/xbook717/files/f4.gif" border="0" align="absmiddle" alt="(!LANG:

The advantage of neural networks over inductive inference lies in solving not only classifying, but also predictive problems. The possibility of a non-linear nature of the functional dependence of output and input features allows you to build more accurate classifications.

The process of solving problems due to matrix transformations is carried out very quickly. In fact, a parallel process of passing through a neural network is simulated, in contrast to the sequential process in inductive systems. Neural networks can also be implemented in hardware in the form of neurocomputers with associative memory.

Recently, neural networks have received rapid development and are very actively used in the financial field. Examples of the implementation of neural networks include:

"System for forecasting the dynamics of exchange rates for Chemical Bank" (Logica);

"Forecasting System for the London Stock Exchange" (SearchSpace);

Investment Management for Mellon Bank (NeuralWare) and others.

NeuroSolution, Neural Works Professional II/Plus, Process Advisor, NeuroShell 2, BrainMaiker Pro, NeurOn-line, etc. should be singled out as tools for developing neural networks.

Case based systems(Case-based reasoning). In these systems, the knowledge base does not contain descriptions of generalized situations, but the actual situations or precedents themselves. Then the search for a solution to the problem is reduced to a search by analogy (inference from particular to particular):

1. Getting detailed information about the current problem;

2. Comparison of the received information with the values ​​of the signs of precedents from the knowledge base;

3. Selection of a precedent from the knowledge base that is closest to the problem under consideration;

4. If necessary, the selected precedent is adapted to the current problem;

5. Checking the correctness of each solution obtained;

6. Entering detailed information about the solution obtained into the knowledge base.

As well as for inductive systems, precedents are described by a set of features, according to which fast search indexes are built. But unlike inductive systems, a search is allowed to obtain a set of feasible alternatives, each of which is evaluated by a certain measure of similarity with the analyzed situation. Usually, as a measure of the similarity of two precedents, a function is taken from the weighted sum of the values ​​of the coinciding features of precedents normalized on some common relative deviation scale. Formally, the complete similarity of SIM between cases A and B, described by p features, can be expressed:

formula" src="http://hi-edu.ru/e-books/xbook717/files/f12.gif" border="0" align="absmiddle" alt="(!LANG:- local similarity (deviation) of the values ​​of the i-th feature of two precedents A and B, normalized on the scale

The following functions can be used as full similarity functions F:

formula" src="http://hi-edu.ru/e-books/xbook717/files/f14.gif" border="0" align="absmiddle" alt="(!LANG:- Minkowski

formula" src="http://hi-edu.ru/e-books/xbook717/files/f16.gif" border="0" align="absmiddle" alt="(!LANG:- Maximum

where p > 0, example ">p number of features, formula" src="http://hi-edu.ru/e-books/xbook717/files/f18.gif" border="0" align="absmiddle" alt ="(!LANG: = 1.

Further, the most suitable solutions are adapted according to special algorithms to real situations. As adaptation methods, the methods of re-concretization of variables, refinement of parameters, replacement of some components of the solution with others are mainly used. Learning the system comes down to memorizing each new processed situation with the decisions made in the database of precedents.

The most well-known application development tools that use precedent search are: CBR-Express (Inference), REMIND (Cognitive Systems), ReCall (Isoft SA), KATE tools (Acknosoft), Pattern Recognition Workbench (Unica), etc. Using these systems you can create various applications for solving problems of diagnostics, risk analysis, prediction, control and training. Case-based systems are used as advanced knowledge dissemination systems or as context-sensitive help systems. For example, the Dell Help Center in Dublin uses the CBR-Express system to help the center respond to more phone calls. Thanks to this system, the productivity of 200 employees of the center has grown to 3,000 calls per day from users. An example of such a consultation may look like this (Fig. 1.15).

Extraction of knowledge based on information repositories(Data mining based on Data Warehouse). Unlike an intelligent database, an information warehouse is a repository of significant information extracted from operational databases, which is intended primarily for operational data analysis (implementation of OLAP technology). The query is formulated as a result of using an intelligent interface that allows you to flexibly define significant analysis features for arbitrary data grouping in the dialog. The most well-known tools for supporting information warehouses and OLAP technologies are such tools as the statistical package of SAS applications, specialized programs Business Objects, Oracle Express, domestic software products PolyAnalyst, Kontur Standard, etc.

Typical tasks of operational situational analysis, solved on the basis of information storages, are:

Determining the profile of consumers of a particular product;

Predicting changes in the market situation;

Analysis of dependences of signs of situations (correlation analysis), etc.

To solve these problems, it is necessary to use methods for extracting knowledge from databases (Data Mining or Knowledge Discovery), based on the use of multivariate statistical analysis methods, inductive methods for constructing decision trees, neural networks, and genetic algorithms.

Let us consider the essence of the application of a previously unconsidered method based on the use of genetic algorithms. Let it be required to determine a set of economic indicators that have the greatest impact on the positive dynamics of market behavior. Then the set of indicators can be considered as a set of chromosomes that determine the qualities of the individual, that is, the solution to the problem. The values ​​of the indicators that determine the decision correspond to the genes.

The search for the optimal solution to the problem is similar to the evolution of a population of individuals represented by their sets of chromosomes. Three mechanisms operate in this evolution: selection of the strongest - sets of chromosomes, which correspond to the most optimal solutions; crossing - the production of new individuals by mixing chromosome sets of selected individuals; and mutations - random changes in genes in some individuals in a population. As a result of the change of generations, in the end, such a solution to the problem is developed, which can no longer be further improved.

Genetic algorithms have a number of disadvantages. The chromosome selection criterion and the procedure itself are heuristic and do not always guarantee finding the best solution. As in real life, evolution can stop at some unproductive branch. On the other hand, one can pick up examples when promising continuations are excluded from evolution by a genetic algorithm. This becomes especially noticeable when solving large-scale problems with complex internal connections. An example of the development of systems based on genetic algorithms is the GeneHunter system from the Ward Systems Group.

The application of methods of intellectual analysis based on information warehouses in practice increasingly demonstrates the need to integrate intelligent and traditional information technologies, the combined use of various methods for representing and deriving knowledge, and the complexity of the architecture of information systems (see paragraph 1.6. and Chapter 7).

In the context of the dynamic development of economic objects, the requirements for the adaptability of information systems to changes are increasing. These requirements are as follows:

IS at any given time must adequately support the organization of business processes.

IS reconstruction should be carried out whenever there is a need to reorganize business processes.

Reconstruction of IS should be carried out quickly and at minimal cost.

Given the high dynamism of modern business processes, we can conclude that the adaptability of IS is unthinkable without intellectualization her architecture. The core of an adaptive IS is a constantly evolving problem area model (enterprise), maintained in a special knowledge base - repository, on the basis of which the generation or configuration of software is carried out. Thus, the design and adaptation of IS is reduced, first of all, to building a model of the problem area and its timely adjustment. Hence, the adaptive system can be attributed to the class of an intelligent information system based on the model of the problem area.

When designing an information system, two approaches are usually used: original or standard design. The first approach involves the development of an information system "from scratch" in accordance with the requirements of the economic object, the second approach is the adaptation of standard developments to the features of the economic object. The first approach, as a rule, is implemented on the basis of the application computer-aided design systems or CASE technologies, such as Designer 2000 (Oracle), SilverRun (SilverRun Technology), Natural LightStorm (Software AG), etc., the second approach is based on the application IC component design systems, such as R/3 (SAP), BAAN IV (Baan Corp), Galaxy (New Atlant), etc.

From the point of view of the adaptability of the information system to the business processes of the economic object, both approaches are guided by a preliminary thorough study of the economic object and its modeling. The difference between the approaches is as follows: when using the repository-based CASE technology, when a change occurs, the software is generated (recreated) every time, and when using the component technology, the programs are configured and only in rare cases they are processed using CASE tools, for example, use of fourth generation languages ​​(4GL).

To model the problem area and subsequent configurations of the information system from individual components (software modules), special software tools are used, for example, R / 3 Business Engineer and BAAN DEM (Dynamic Enterprise Modeler). The undoubted advantage of using model-oriented component systems, such as R / 3 or BAAN IV, over CASE technologies is the accumulation of experience in designing information systems for various industries and types of production in the form of standard models or so-called reference / reference (reference) models, which are supplied with the software product in the form of a filled repository. Thus, along with the software product, users acquire a “know-how” knowledge base on effective methods for organizing and managing business processes that can be adapted in accordance with the specifics of a particular economic object.

In a generalized form, the configuration of adaptive information systems based on component technology is shown in fig. 1.16.

The repository base model contains a description objects, functions (operations), processes (set of operations), which are implemented in the program modules of the component system. At the same time, the task rules (business rules) maintaining the integrity of the information system, which establish the conditions for checking the correctness of the joint application of business process operations and the software modules supporting them. Thus, the diversity and flexibility of defining business processes and the corresponding configurations of the information system is specified using a set of business rules.

Typical models correspond to typical information system configurations made for certain industries (automotive, electronics, oil and gas, etc.) or types of production (individual, serial, mass, continuous, etc.).

Enterprise model (problem area) is built either by linking or copying fragments of the main or generic models in accordance with the specific features of the enterprise, for example, as in the BAAN Orgware tool, or as a result of viewing these models and an expert survey, as in the R / 3 Business Engineer tool. Moreover, in the latter case, the user is prompted to determine the values ​​of not all parameters, but only those that are interconnected in the context of the dialog and are described by business rules.

The generated enterprise model in the form of a meta description is stored in the repository and, if necessary, can be corrected. Further, according to the formed enterprise model, the configuration of the information system is automatically carried out, during which semantic control is performed according to the corresponding business rules.

The disadvantages of the described information system configuration scheme is the lack of means for evaluating the enterprise model. In order to be able to choose the optimal options for the configuration of the information system, as a rule, tools for exporting the model to external modeling systems are used. Thus, the R/3 system provides for the export (import) of models to (from) the environment (s) of the ARIS Toolset tool, which allows you to perform a functional cost analysis of the effectiveness of the modeled business processes and their dynamic simulation.

The continuous changes taking place in the economy dictate the need for constant updating of the knowledge of enterprises and organizations as an intellectual capital that ensures sustainable strategic positions of enterprises in the market. According to B.Z. Milner, “a new management function is being formed, the task of which is to accumulate intellectual capital, identify and disseminate existing information and experience, and create prerequisites for the dissemination and transfer of knowledge. It is knowledge that becomes the source of high productivity, innovation and competitive advantage.” The new knowledge management function is implemented as a set of processes for the systematic acquisition, synthesis, exchange and use of knowledge within the organization. Knowledge management is characterized by the collective formation and use of both internal and external sources of knowledge (information resources).

According to a survey of Fortune 1000 CEOs, 97% of managers said there are processes that are key to the company that could be significantly improved if only more employees knew about them. In the same survey, 87% of its participants say that costly errors occur precisely because employees do not receive the necessary information in time.

The quality of the knowledge used directly affects the efficiency of the following business processes:

Making managerial decisions in strategic, tactical and operational management as a result of obtaining timely access to relevant knowledge;

Innovative activity due to the possibility of collective formation of ideas and reducing the cost of duplication of work, ensuring the acceleration of the innovation cycle;

Continuous professional development of employees of enterprises in real time;

Providing partners (suppliers, contractor, customers) in addition to their core services access to accumulated knowledge, including consulting and training.

Under knowledge management system (KMS) we will understand the set of organizational procedures, organizational units (knowledge management services) and computer technologies that ensure the integration of heterogeneous sources of knowledge and their collective use in business processes.

According to Popov E.V., the world expenditures on the creation and operation of the CPS, which amounted to $ 2 billion in 1999, will increase to $ 12 billion by 2003, in which $ 7.9 billion will be associated with the provision of services, $1.8 billion for software, $0.9 billion for infrastructure maintenance, and $1.4 billion for internal resources. However, according to Delphi Consulting Group experts, currently only 12 percent of corporate knowledge is formalized in computer databases and knowledge.

A distinctive feature of the knowledge management system is the integration of many heterogeneous, often geographically distributed sources of knowledge to solve common problems. KMS integrates knowledge from both internal and external sources. Knowledge sources can have an undocumented form (implicit knowledge of experts), a documented textual, tabular, graphical form, and a structured form in the form of knowledge bases of expert systems.

Internal sources of knowledge include:

Technical documentation, description of production and business processes,

Intra-company databases (data bases) and information warehouses (data warehouse),

Knowledge bases of the work experience of specialists (“best practice”),

Description of knowledge profiles of specialists (experts),

Specialized expert systems.

External sources of knowledge include:

Materials of publications and news contained in the INTERNET,

Electronic learning systems,

External databases of partners and statistical databases in the regional, product and industry sections,

Directory of experts and consulting companies specializing in specific problem areas, links to forums on the INTERNET,

Reference models for organizing business processes (industry and standard solutions).

A knowledge management system is usually used in two aspects:

Providing high-quality knowledge of the processes for solving various problems.

Creation of an interactive environment for the interaction of specialists in the process of solving problems.

From the point of view of the first aspect, KMS, unlike traditional information systems for documentation support (information retrieval systems), turns knowledge into a finished product with a high use value, since knowledge, unlike a set of all kinds of information related to a request, exactly corresponds to the nature of the problem being solved and can be used directly. when developing a solution.

From the point of view of the second aspect, KMS creates an interactive environment for people to communicate, in which the creative ability to generate new knowledge is increased, which immediately enters the corporate memory for subsequent use. With the help of KMS, any enterprise or organization turns into a learning organization, creating a “knowledge spiral”, in which “unknown (implicit) knowledge must be identified and disseminated in order to become part of the individualized knowledge base of each employee. The spiral is renewed each time to rise to a new level, expanding the knowledge base applied to different areas of the organization.

Thus, in order for the process of updating knowledge to be constant, it is necessary to create constantly functioning knowledge management systems that could not only combine individual sources of knowledge of individual users, but also extract knowledge from external sources of knowledge, statistical databases, and the INTERNET information space. This requires the ability to connect the corporate knowledge management system to other knowledge management systems based on common approaches to the conceptualization of knowledge.

Knowledge management systems are characterized by the following features:

Intelligent assistance. A knowledge management system, unlike expert systems, does not replace an expert in the process of solving problems, but helps him by providing relevant information and decision-making rules in a particular situation. At the same time, in the process of solving the problem, the user considers various solutions presented by the knowledge management system, modifies the problem statement or models the situation, thus choosing the most appropriate solutions. There may be another mode of solving the problem, when the user independently solves the problem, and evaluates the result of the solution using the knowledge management system for correctness and efficiency, for example, by contacting expert colleagues or expert systems using the knowledge management system.

Collection and systematic organization of knowledge from various sources. The integration of many heterogeneous sources of knowledge is carried out on the basis of a single system of conceptualization of knowledge. The main requirement for knowledge sources is to prevent the loss and increase the availability of all types of corporate knowledge by providing a centralized, well-structured information repository. The structuring of the information warehouse involves the creation and description of a unified knowledge system based on a taxonomy of conceptual concepts, a meta-knowledge base or ontology, through which you can access various sources of knowledge.

Knowledge base design minimization as filling the conceptual scheme. Not all sources of knowledge may be known, or they may be inconsistent or change frequently. The knowledge management system, on the one hand, should provide methods for handling such situations, and on the other hand, provide easy connection of new sources of knowledge as they appear. At the same time, the conceptual scheme of the knowledge management system itself should be modified.

Rapid adaptation of the knowledge management system to changing information needs. Adaptation of the knowledge management system is carried out on the basis of users' feedback to the system as a result of their formulation of new requirements and analysis sections, as well as informing about knowledge gaps, inaccuracies and delays. In addition, the base of cases is filled with indication of successful and unsuccessful problem solving by users. This transforms individual knowledge into group knowledge available to other users of the knowledge management system. The form of collected and distributed knowledge becomes generative and reusable.

From this point of view, it is useful to accumulate a knowledge base of precedents for solving problems, to generalize typical situations and errors, and to disseminate the generated knowledge.

Integration with existing software environment. A variety of analytical tools are connected to the knowledge management system, which allow the extraction of implicit information contained in knowledge sources using methods such as statistical analysis and neural networks, expert systems, mathematical and simulation modeling. These tools allow you to detect patterns in the reflected reality and identify the most rational behavior in existing conditions. The user becomes an experimental researcher who not only asks for information of interest to him, but also puts forward and tests various hypotheses. In this sense, it is useful to connect specialized software tools.

Active presentation of relevant information. The knowledge management system becomes a competent partner in cooperative problem solving, taking into account the range of information needs of the user and forming knowledge for him, based on his supposed interests. This property is especially effective in customer relationship management (CRM) systems.

The integration of heterogeneous sources of knowledge, the interdisciplinary nature of their use, the need to attract external sources of knowledge, the exchange of knowledge between users involves the development knowledge management system architecture based on a common information space in the form of an integrated memory, which can be represented at three interacting levels:

Object level- storage of knowledge sources annotated with the help of a specially developed system of categories of knowledge sources and their indexing.

Conceptual level- definition of a conceptual model of the knowledge structure (a system of categories), common to all sources of knowledge, that is, the development of an ontology.

Application layer- determination of the goal and restrictions on the solution of an intellectual problem by the user, that is, setting its global and local context. In this case, the user's knowledge profile must be defined.

Let's consider the listed levels of corporate memory organization in more detail. When describing the organization of knowledge at these levels, examples of specific operating KMS are given, described in .

Enterprise Memory Object Level

Possible sources of knowledge that are connected to the KMS are presented in Table 1.2.

Table 1.2

Sources of knowledge

Document without a title

Sources of knowledge	Knowledge Acquisition Methods	Knowledge Examples
People and groups	Documentation, Structuring, Connection	Examples of situations, Rules, objects Knowledge profile
In an implicit form, by directly connecting specialists and experts to a computer network by describing their knowledge profile. Experience in maintaining best practice cases is known, for example, at Huges Electronics, part of General Motors, which maintains a database of the best plant refurbishment projects. Each project is associated with a short description and information to contact the responsible persons. Statistical databases and information warehouses are a source for extracting implicit knowledge using data mining methods: inductive construction of decision trees, cluster and regression analysis, and construction of neural networks. Information warehouses can contain huge amounts of data. For example, Chase Manhatten Bank has over 560 GB of storage, MasterCard OnLine has 1.2 TB. Specialized tools can be used to collect statistical data. For example, to extract knowledge from financial information in the EDGAR (Electronic Data Gathering and Retrival System) system, the EdgarScan system was developed, which operates in the Internet environment. Text sources of knowledge are connected to the KMS using the technology of filtering text messages analyzed in the knowledge source based on a list of topics that define the taxonomy hierarchy of terms in a particular subject area. As a rule, filtering is carried out by categories and priorities of importance. For example, the Odie (On demand Information Extractor) system scans about 1,000 breaking news articles every night to extract knowledge about management trends. Odie, designed for US and European news reviewers, uses the recognition of stylized phrases in business news articles and knowledge of syntax rules to recognize relevant business events. Graphic diagrams of reference models are stored in special repositories that describe meta-information about the organization of business processes. For example, the SAP R/3 repository contains about 100 proven business cases and models for various industries. With the help of expert rules, reference models can be transformed into specific models for organizing business processes of enterprises. Conceptual level knowledge structure of corporate memory The basis of the conceptual level of the knowledge structure is the taxonomy of the concepts used (ontology), designed to identify the various components of knowledge. Ontology can be viewed as a subject area rubrication system, which integrates heterogeneous sources of knowledge. On the other hand, the ontology is considered as a thesaurus dictionary shared in the KMS to facilitate user communication, formulating and interpreting their queries. The conceptual level of corporate memory provides a semantic interpretation of queries to the KMS, which implements a unified intelligent access to a variety of knowledge sources. This results in the following key benefits: Accurate and efficient KMS access to knowledge sources relevant to the context of the task (specific situation). Better understanding and interpretation by the user of the acquired knowledge in this context with the possibility of additional reference calls to corporate memory. Information modeling, that is, iterative refinement of information needs in the process of solving a problem. As an example of the use of thesaurus dictionaries, we can cite the industrial information retrieval system Retrieval Ware (Canbera), which allows using a regular web browser to perform full-text and attributive search in various knowledge sources: electronic archives of text documents, web-pecypcax, formatted data bases, spreadsheets, graphic, sound and visual images (in 250 formats). The semantic network reflected in the dictionary-thesaurus for the English language includes 400,000 words and more than 1,600,000 links between them. The dictionary of the localized version of the Russian Semantic Server software product currently contains about 42 thousand words and idiomatic expressions. Semantic search technology also allows you to use several dictionaries at the same time. For example, along with the basic dictionary, the system can use an industry dictionary, an internal organization dictionary, and a user's personal dictionary, which can be developed as needed. The limitation of the thesaurus organization of knowledge is associated with a narrow focus on solving only search problems. To expand the range of KMS tasks related to the choice of alternatives in justifying projects, making decisions, and training, a more complex organization of knowledge is required, which involves defining the logic of solving problems in a declarative form or creating an ontology of tasks. A good example of this kind of system is the Ontolingua system. The Ontolingua system is designed to support the formal specification of user tasks based on a library of formal descriptions of task fragments, models and concepts, as well as to maintain the library of fragments itself (Fig. 1.17). Otologies can be used not only in knowledge management systems, but also in transactional systems such as e-commerce systems, logistics systems, virtual enterprises, which require multi-agent knowledge exchange technology. Application layer Intellectual tasks that are solved on the basis of the CPS are characterized by weak formalization, which implies the fuzziness of setting goals for solving the problem and describing the conditions for solving the problem. In addition, the level of knowledge and the system of criteria for evaluating a solution may differ for different users. Typically, a typical solution to an intellectual problem comes down to the following steps: Formulation of the problem statement. Selection of knowledge sources relevant to the task. Understanding the selected material (training, consultation). Problem solving (development, solution configuration). Checking the admissibility of the solution to the problem (evaluation of the solution to the problem, consultation with experts, expert systems). Decision making and monitoring of its implementation. Storing the results of solving a problem in corporate memory. At each step of solving an intellectual problem, the CMS can be used, the work of which is reduced to an iterative series of searches in the corporate memory, which ensures the accuracy of achieving the goal of each stage. For example, an enterprise solves the problem of choosing a strategy for increasing production efficiency, which belongs to the class of poorly formalized tasks. To begin with, it is necessary to request information on possible goals, criteria and methods for improving production efficiency. Next, describe your business. Based on the formulated task, the system will select the sources of knowledge. The study of the material can cause a clarifying or explanatory dialogue. The adopted decision can be sent to specialists selected with the help of the CMS for examination. In the process of implementing the solution, information about individual steps is entered into the knowledge base to adjust the strategy based on the identified deviations and for use in solving similar problems in the future. At each stage of solving an intellectual problem, the requirements for input and output, the methods and means used, as well as the resources used (specific performers, time, material and cost constraints) are determined and looks like filling out some screen form, in which the following are fixed: The overall purpose of the activity. Context information known from the state of the process or set manually at this step. The screen form, as a rule, is filled with the help of lists of keywords associated with ontology headings. Thus, the screen form of the request reflects the current local context of the search, which determines the features of the next step in solving the problem. One-time access to knowledge in modern KMS comes down mainly to either the search for relevant knowledge by end users, or the dissemination of knowledge among users by the knowledge management system. Thus, both people and KMS can act as initiators of knowledge management processes. On the other hand, the carriers of knowledge addressed by the process initiators can also be both people and KMS. Possible options for interaction between people and KMS are shown in Table 1.3. Table 1.3 Access to knowledge sources Document without a title Linking people in the CPS is carried out using multi-agent technology based on a repository of human skills. Each participant in the knowledge management process in a computer network may have a computer twin - an intelligent agent that enters into negotiations with similar agents in the network to solve a common problem. An example of such a multi-agent system is the ContactFinder system, which searches for experts on the network using bulletin board technology, which is available for messaging between agents. At the same time, the role of ontology is great, as a dictionary of communication between agents and a description of the profile of their knowledge. Linking people and KMS for the selection of relevant knowledge is carried out with the help of search engines in knowledge repositories (“pool-stores”). The lack of intelligence of existing search engines leads to the irrelevance of the selected knowledge. Currently, ontology-based search engines are being developed, for example, the SHOE system, OntoBroker, etc. Special tools, for example, Perspecta and InXight, can be used as graphical tools for visualizing a knowledge map when navigating access paths. Linking BMS and people is carried out according to the “push” technology (push technology), according to which the KMS studies the information needs of users and, in accordance with them, independently distributes or delivers knowledge to consumers. For example, InfoFinder learns the interests of users by the sets of generated messages or requested documents. In addition, InfoFinder uses heuristics to collect additional, more accurate information about user queries, allowing InfoFinder to find documents by anticipating user queries. Linking BMS and BMS in the process of implementing more complex requests, when one CMS cannot cope with the task and turns to another CMS for help, as a rule, it is performed using multi-agent technology. In addition, dedicated knowledge guides can be used to facilitate navigation of knowledge selection paths in an integrated knowledge space, such as CoopersLibrand's "Michelin Guides" to help understand the content and location of knowledge sources. Andersen Consulting maintains a central repository of knowledge maps that links various sources of knowledge. What to remember Intelligent Information System (IIS) is an IS based on the concept of using a knowledge base to generate algorithms for solving economic problems of various classes, depending on the specific information needs of users. The most important features of the classification of IIS: developed communication skills, complexity (poor formalizability of the algorithm), self-learning ability, adaptability. The main subclasses of IIS: intelligent databases, incl. with interfaces using natural language, hypertext and multimedia, cognitive graphics; static and dynamic expert systems; self-learning systems based on the principles of inductive inference, neural systems, search for precedents, organization of information storages; adaptive information systems based on the use of CASE-technologies and/or component technologies, knowledge management systems. Intelligent Interface System- this is an IIS designed to search for implicit information in a database or text for arbitrary queries, usually compiled in a limited natural language. Expert system (ES)- this is an IIS designed to solve weakly formalized tasks based on the experience of experts in the problem area accumulated in the knowledge base. Participants in the development and operation of ES: experts, knowledge engineers, users. An expert is a specialist whose knowledge is stored in the knowledge base. Knowledge Engineer- a specialist who is engaged in the extraction of knowledge and its formalization in the knowledge base. The user is a specialist whose intellectual abilities are expanded through the use of ES in practice. The main components of the ES architecture: knowledge base, inference mechanisms, explanations, knowledge acquisition, intelligent interface. The knowledge base is the central component of the ES, which determines the value of the ES and with which the main development costs are associated. The knowledge base is a repository of knowledge units that describe the attributes and actions associated with the objects of the problem area, as well as the possible uncertainties. A unit of knowledge is an elementary structural unit (description of one object, one action), which has a complete meaning. Rules and/or objects are usually used as units of knowledge. Knowledge Uncertainty- this is either incompleteness, or unreliability, or ambiguity, or a qualitative (instead of quantitative) assessment of a unit of knowledge. An inference mechanism is a generalized procedure for searching for a solution to a problem, which, based on the knowledge base and in accordance with the information need of the user, builds a chain of reasoning (logically connected units of knowledge) leading to a specific result. deductive inference(from general to particular) - derivation of particular statements by substituting other well-known particular statements into general statements. Distinguish between direct (from data to target) and reverse (from target to data) chains of reasoning (argumentation). inductive output(from particular to general) - conclusion (generalization) based on a set of particular statements of general statements (from examples of real practice of rules). Abductive inference(from particular to particular) - derivation of particular statements based on the search for other similar statements (precedents). Knowledge Acquisition Mechanism- this is a procedure for accumulating knowledge in a knowledge base, including input, control of the completeness and consistency of knowledge units and, possibly, automatic derivation of new knowledge units from the input information. Explanation mechanism is a procedure that justifies the result obtained by the inference mechanism. Intelligent Interface is a procedure that interprets the user's request to the knowledge base and generates a response in a form convenient for him. Purpose of the expert system: advising and training inexperienced users, assisting experts in solving problems, advising experts on issues from related fields of knowledge (integration of knowledge sources). Static expert system- this is an ES that solves problems in conditions of initial data and knowledge that do not change in time. Dynamic Expert System- this is an ES that solves problems in the conditions of initial data and knowledge changing in time. Analytical expert system- this is an ES that evaluates options for solutions (testing hypotheses). Synthetic expert system- this is an ES that generates decision options (formation of hypotheses). Classes of tasks to be solved in an expert system: interpretation, diagnostics, forecasting, design, planning, monitoring, correction, management. Self-learning system is an IIS that automatically generates units of knowledge based on examples of real practice. System with inductive output is a self-learning IIS, which builds decision trees based on learning from real practice examples. A neural network is a self-learning IIS, which, based on learning from real practice examples, builds an associative network of concepts (neurons) for parallel search for solutions on it. Case based system is a self-learning IIS that stores the actual decision precedents (examples) as knowledge units and allows, upon request, to select and adapt the most similar precedents. Data Mining- this is a set of methods for automatically extracting knowledge from specially organized information storages (Data Warehouse), which include statistical methods, inductive inference, neural networks, genetic algorithms. Adaptive Information System- this is an IIS that changes its structure in accordance with a change in the model of the problem area. Problem domain model- reflection of the structure of objects, functions, processes, rules associated with the functioning of the problem area. A repository is a repository of meta-knowledge about the structure of factual and operational knowledge or a problem domain model. Case technology is a technology that allows you to generate an information system based on a model of a problem area stored in a repository. Component Technology- a technology that allows you to configure an information system from ready-made standard components based on the model of the problem area stored in the repository. Knowledge Management System (KMS)- an interconnected set of organizational procedures, people and information technologies, which ensures the collection, accumulation, organization, dissemination and use of knowledge to solve the problems of high-quality information service (ensuring) the implementation of business processes and interactive interaction of specialists. Enterprise Memory- a repository of knowledge sources and their meta-descriptions for collective use in the organization. KMS Knowledge Sources- the experience of specialists, presented in the form of precedents, structured knowledge bases of expert systems, descriptions of knowledge profiles; documentary sources of knowledge inside and outside the organization; databases and storages of formatted data; reference models for organizing business processes. Ontology is a conceptual description of the knowledge structure for formalized and non-formalized knowledge sources. KMS applications - search and use of knowledge to solve intellectual problems of decision justification, design, innovation; education; exchange of knowledge in the process of interaction between specialists; dissemination of knowledge for familiarization in the potential plan.

Lecture

Topic: "Intellectual technologies and systems"

Plan:

1. The concept of artificial intelligence. Intelligent Information

technology.

2. Classification of intelligent information systems.

3. Expert systems as the main type of intelligent systems.

4. Artificial neural networks.

The use of information technology (IT) in various areas of human activity, the growth of information volumes and the need to respond quickly in any situation required the search for adequate ways to solve emerging problems. The most effective of them is the way of intellectualization of information technologies.

Question number 1 The concept of artificial intelligence.

Intelligent Information Technology

The new information technology is based primarily on intellectual technologies and the theory of artificial intelligence.

The term intellect comes from the Latin intellectus - which means mind, reason, reason; human thinking ability.

Under artificial intelligence understand the ability of computer systems to act intelligently. Most often, this refers to the abilities associated with human thinking.

Artificial intelligence- a section of computer science related to the development of intelligent programs for computers.

Artificial intelligence (AI) is a scientific direction that emerged at the intersection of cybernetics, linguistics, psychology and programming.

Under intelligent information technology understand such information technologies, which provide the following capabilities:

• the presence of knowledge bases that reflect the experience of specific people, groups, societies, humanity as a whole, in solving such problems as: decision making, design, extraction of meaning, explanation, training;
• availability of thinking models based on knowledge bases: rules and logical conclusions; argumentation and reasoning; recognition and classification of situations; generalizations and understanding, etc.;
• the ability to form quite clear decisions based on fuzzy, incomplete, underdetermined data;
• the ability to explain conclusions and decisions, that is, the presence of an explanation mechanism;
• the ability to learn, retrain and, consequently, to develop.

History of intelligent information technology

Let us turn to the history of the development of IIT, which dates back to the 60s of the last century and includes several main periods.

• 60-70s. These are years of awareness of the possibilities of artificial intelligence and the formation of an order to support decision-making and management processes.
• 70-80s. At this stage, there is an awareness of the importance of knowledge for the formation of adequate decisions; EXPERT SYSTEMS appear.
• since the 80s Until now. There are integrated (hybrid) models of knowledge representation that combine the following types of intelligence: search, computational, logical and figurative. Creation of neural networks

A feature of intelligent information technologies (IIT) is their "universality". They have practically no restrictions on applications in areas such as management, design, machine translation, diagnostics, pattern recognition, speech synthesis, etc.

IIT is also widely used for distributed complex problem solving, collaborative product design, building virtual enterprises, modeling large manufacturing systems and e-commerce, electronic development of complex computer systems, managing knowledge and information systems, etc. Another effective application is information retrieval. in the Internet and other global networks, its structuring and delivery to the customer.

Question number 2 Classification of intelligent information systems

IIS is characterized by the following features:

Developed communication skills (the way the end user interacts with the system);

The ability to solve complex, poorly formalized problems that require the construction of an original solution algorithm depending on the specific situation, characterized by uncertainty and dynamism of the initial data and knowledge;

Ability to self-learning, i.e. the ability of the system to automatically extract knowledge from accumulated experience and apply it to solve problems;

Adaptability is the ability of the system to develop in accordance with objective changes in the field of knowledge.

Each of the listed signs conditionally corresponds to its own class of IIS.

1. Systems with an intelligent interface (communicative abilities):

- Intelligent databases. They allow, unlike traditional databases, to provide a selection of the necessary information that is not explicitly present, but derived from the totality of stored data.

- Natural language interface . It is used for access to intelligent databases, contextual search of documentary textual information, voice input of commands in control systems, machine translation from foreign languages.

- Hypertext systems. Used to implement keyword search in databases with textual information.

- Context help systems . They belong to the class of knowledge dissemination systems. Such systems, as a rule, are appendices to the documentation. In these systems, the user describes the problem, and the system, based on an additional dialogue, concretizes it and searches for recommendations on this problem.

- Cognitive graphics systems . They are focused on communication with the IIS user through graphic images that are generated in accordance with the change in the parameters of the simulated or observed processes. The use of cognitive graphics is especially important in monitoring and operational management systems, in teaching and training systems, in real-time operational decision-making systems.

2. Expert systems(solution of complex poorly formalized problems). They are used to solve non-formalized problems, which include tasks that have one of the following characteristics:

Tasks cannot be represented in numerical form;

Initial data and knowledge about the subject area are ambiguous, inaccurate, inconsistent;

Goals cannot be expressed in terms of a well-defined objective function;

There is no unique algorithmic solution to the problem;

The main difference between ES and SII from data processing systems is that they use a symbolic rather than a numerical way of representing data, and logical inference and heuristic search for solutions are used as methods of information processing.