Principles for the development of intelligent interfaces

The problem of developing large-scale computer networks relates to NP-complete problems that cannot be solved only by mathematical methods. Consequently, the computer network design system should support hybrid knowledge representation models, work with unclearly presented and multi-format data, that is, be open. Moreover, the system should be able to communicate with the user in a language close to natural.

The use of hybrid knowledge representation models, that is, models that use symbolic and neural network knowledge representations, has a number of important advantages: firstly, it is possible to use the widest range of expert knowledge about the subject area in the intellectual system, secondly, knowledge is organized between various modules of the intellectual system (including between modules that use different paradigms of representation and obtaining expert knowledge).

Difference from existing developments is openness for adding modules with various data processing methods. The hybrid model of knowledge representation allows to describe the solution of a complex problem as an interconnected set of simpler subtasks, for each of which it is possible to use different paradigms of knowledge representation. As a result, the decision support system (DSS) can be a complex structure of interconnected neural networks, fragments of symbolic databases, statistical models, etc.

The openness of the system allows you to connect the modules and to convert data from external multi-format databases into the internal format of the program to supplement the knowledge base, which makes the system flexible and independent of the format of incoming data from outside.

The structure of the knowledge base is defined as a directed graph whose nodes represent the data processing modules, and the edges define the direction and sequence of the subtask solution.

The classic way to describe graph models is to use incident matrices and adjacency matrices. However, in this case the problem of machine representation of the graph is complicated by the fact that along with the structure of the graph it is necessary to store the description of a separate task.

To represent the structure of knowledge in the theory of artificial intelligence systems, a language of semantic networks has been developed. However, for describing hierarchically decomposed tasks, such an approach may turn out to be redundant, since tasks are connected by homogeneous arcs. In addition, the inference procedure for the tree graph is much simpler than the inference procedure for the semantic network.

The basis for the development was the model used in the Analytic system [2]. A distinctive feature of the graph model is the possibility of using several paradigms of knowledge representation in solving one problem. For a separate task several knowledge bases can be formed. They will have a common target parameter (group of parameters), but they can use different sets of input data to solve.

User and system communication - dialogue in a limited natural language (NL). NL-dialog is a separate DLL module with a multilevel linguistic processor (LP), which consists of five blocks: a lexical analyzer, a block for defining words, numbers and constants, a block of translation into an internal formal language, a block of translation into a formal query language to the knowledge base and block direct request. The intensive development of information technologies in Russia, the need to exchange vast amounts of information within and between enterprises, criticality to the speed of data transmission and their reliability make the problem of designing corporate networks one of the top priorities and provide great interest in this development.