The article investigates the problem of automated development of basic ontology. A method, algorithm and means for extracting knowledge from natural text are proposed. It is shown that such an algorithm should be multistage and include a hierarchical multi-level procedure for recognizing concepts, relationships, predicates and rules, which are introduced as a result of ontology. The analysis of the subject area is the search and analysis of various information systems analogues. The analysis of methods and criteria of information systems is carried out. The analysis of the system and its functionality is presented. This paper examines methods and models of intellectual text processing, the results of which are intended to build software ontologies, and are used during Ontology Learning, when it is necessary to improve, extend, modify an existing ontology model, or build ontology from basic ontology, having only textcollection collections as sources of knowledge. In the latter case, the task is particularly complex and requires the use of the full range of text mining methods (Text Mining as TM). The paper deals with the solution of TM problems at different stages of the PMT processing: obtaining information (identifying entities - concepts and terms, their properties, facts, events, establishing relationships between entities, in particular associative ones), categorization, and clustering, semantic annotation. Below we will consider the tools of automated analysis of natural language and software products implemented on their basis for filling the system.
Keywords: ontology, ontology training, automated development, knowledge base, text document, content classification. 1
This paper examines methods and models of intellectual text processing, the results of
which are intended to build software ontologies, and are used during Ontology
Learning, when it is necessary to improve, extend, modify an existing ontology model, or
build ontology from basic ontology, having only text-collection collections as sources
of knowledge. In the latter case, the task is particularly complex and requires the use
of the full range of text mining methods (Text Mining as TM) [
There are a number of perspective linguistic developments, among which it is
expedient to single out the method of Part-of-Speech-tagging, which consists in the
automatic recognition of which part of the language belongs to each word in the text
[
Unlike lexical-grammatical analysis, the purpose of syntactic parsing (Text
Parsing) is the automatic construction of a phrase tree that is, finding interdependencies
between different levels of a sentence. There are a number of different approaches to
parsing, for example, Ergo Linguistic Technologies Parser, developed by D.
Bickerton and F. Braalik of Honolulu University [
Obviously, it takes a lot of time and resources to manually build a complete related ontology for a specific subject area. The reason for this cost is that such ontologies must contain tens of thousands of elements in order to be capable of solving the wide range of applications that arise in this software. Therefore, the manual construction of ontology by a human operator is a lengthy routine process that, in addition, requires a thorough knowledge of the subject area and an understanding of the principles of ontology construction.
Therefore, we will build mathematical support for the automation of ontology construction, and more precisely its construction. Because we believe that the basic terms and the relationship between them should be entered manually by an expert person into the ontology. We call this ontology a basic one and denote it Obase Cb , Rb , Fb . That is, the construction of ontology begins from the moment when it already has some data. Therefore, we will call this process the development of a basic ontology and denote: : Obase O . In order to build ontologies that adequately describe semantic software models, it is necessary, first of all, to solve the problems of obtaining knowledge from different sources in order to identify many concepts and establish a hierarchy on that set. Since much of the information is contained in natural-language texts, it is promising to acquire knowledge of textual information as well as intellectually processing specially selected collections of natural-language texts. 3 3.1
One of the most effective approaches to completing ontology is its automated teaching of natural texts. Automated filling can be implemented by analyzing text documents using a knowledge processor (Fig. 1).
Text
Linguistic processor
Ontology processor
Text
ontology
In the presented scheme, the task of the linguistic processor is to perform its lexical,
lexical, grammatical, syntactic and semantic analysis. As a result, the ontology is
replenished with: concepts, threesomes (subject - action - object) and cause and effect
relationships between SDO threes. The other part of the important relationships
between concepts and their properties is established by the ontology processor, which
builds the ontological structure for each C concept obtained after analyzing the text.
The work of the ontology processor is supported by an appropriate knowledge base,
the main components of which are, firstly, a set of rules, and secondly, a universal
WordNet-type MMWW database [
Therefore, from the point of view of the process of building an ontology, it is necessary to build its term system OT and nomenclature ON . In our approach, the basic ontology should accurately integrate part of the term system (Fig. 2), that is OT and nomenclature ON In our approach, the basic ontology should accurately integrate part of the term system (Fig. 2), that is
OB OT (1) Encyclopedias, the terminological and explanatory dictionaries on which the software terminology system is built, are usually clearly structured and consist of dictionary entries. Therefore, it is necessary to investigate their possible structures in order to recognize the concepts and relationships between them.
textbooks, monographs, ...
Ontology The term system Basic ontology
Nomenclature
Encyclopedias,
glossaries, ...
Building a nomenclature is more complicated. If terms are already highlighted in
dictionaries, then in scientific texts (textbooks, monographs, etc.), they should be
highlighted, search for properties of concepts and relations between concepts.
Methods should be developed for obtaining terms from sources of knowledge, fixing them
and dividing them into categories. Such extraction is related to natural-language
processing of information. Such methods are constantly being improved and are in
constant development, so their submission should be declarative, since such a
representation provides the easiest way to improve it. Scientific texts are monologic. So, we
conclude that technology is needed, which would allow almost in the automatic mode
to create methods of native-language processing of scientific text, which already in
the automatic mode would allow to build ontological models. Thus, in [
The starting point when creating any model of knowledge about software is the choice
of its categorical apparatus. For any abstract systems that use the same thesaurus or
dictionary, there is no guarantee that they will be able to use the same information
correctly until a single conceptualization is adopted. The conceptualization is based
on the category of abstractions that are associated with the construction of the term
that underlies any ontology. We substantiate the construction of the term construction
as a sign of the semiotic system. Today, there is no single correct way to model
software. However, there are some fundamental rules for the development of ontology:
Effective resolution always depends on the proposed program and the expected
extensions;
Ontology development is a must-have iterative process;
Concepts in ontology must be close to objects and relationships in software.
The paper uses a categorical apparatus derived from the work of linguists, logicians,
and computer scientists [
Definition is the union of forms of structural and substantive definitions, in which structural information implies the representation of the most probable niche substantive and from the substantive information the most probable structural interactions of elements of the field of the terms system, so that together these two aspects provide a representation of its integrity and functional validity. This means that it is necessary to have as a substantive definition of the term a verbal definition of the term, and as a structural definition a fragment of a network of signs.
The referent is the representation of the denotation of real world entities (object, phenomenon, process), knowledge described in the sign system. A concept is the knowledge that is expressed in this concept in the conceptual modeling of software.
Intensive is the content of a concept that corresponds to the structural definition and is described as an internal form of the concept that combines its lexis and logo and sufficiently to define the extension. Extension is a concept scope.
Conceptual objects are divided as follows: Essence is tangible and intangible objects, ways of considering them; Property is quantitative, qualitative, relational (ratio); Action is operations, processes, states; Quantities are time, space… Conceptual relationships: Quantitative (coinciding with the theoretical-multiple relations of identity, inclusion, deletion, intersection, union); Qualitative (hierarchical and functional).
In the AI industry, the real world is considered to be objects. Objects can be made up of parts. Objects have properties that matter. Objects can be different in relation to each other. Properties and relationships change over time. At different points in time, events occur that trigger the processes in which objects are involved and change over time. Events can trigger other events, that is, have an effect. The world and its objects can be in different states. 3.3
Methods of construction of ontologies can be divided into groups. The first group will
include traditional methods of natural language processing, and the second will
include methods that relate directly to ontology construction. Consider the technology
of analysis of naturalistic text and construction on their basis ontology, proposed in
[
The implementation methods of the first four blocks are considered to be the most elaborated. However, it should be noted that the studies are ongoing because a satisfactory result of their work has not yet been obtained. Yes, we suggest using language ontology to perform relevant analyzes. The pre-processing function includes lexical analysis, splitting of complex sentences into simple sentences, division of sections, sub-sections, sentences in the source text, verification of compliance with accepted restrictions. At this point, the research of scientific texts is considered inadmissible as complex sub-sentences that combine recursively-embedded meaning sentences.
Knowledge base of linguistic support
MA Production Rules SynA Production Rules
SP Production Rules SemA Production Rules
The set of natural language scientific texts Pre-processing Morphological
analysis
The main task of lexical analysis is to split the input text of a document, which is a sequence of single characters, into a sequence of tokens. From this point of view, all characters in the input sequence are divided into characters that belong to any tokens, and characters that separate the light-seven (delimiters). In some cases, the tokens may not be delimited. As a result of lexical analysis, a set of tokens is formed L = {li |i=1, …, k, k is the number of tokens in the text}. Each token is assigned a vector: i pi , nil , nis , nip , nid , nic (2) where pi is unique token vector number; nil is the sequence number of the tokens in s p the sentence; ni is the order number of the sentence in the text; ni is paragraph number; nid is section number; nic is chapter number.
The main function of verbal morphological analysis is to determine the language part of the token li and assigning it a vector of morphological information i .The analysis of lexemes uses dictionaries of endings, a dictionary of inflectional classes, a dictionary of ready word forms and a dictionary of basics, tables of compatibility of the basics of a inflected class and vectors of morphological information. 3.4
All syntactic units of a naturalistic sentence must be uniquely identified during parsing. Syntax units are constructs in which their elements (components) are joined by syntactic links and relations. Syntax is an expression of the interrelation of elements in a syntactic unit, that is, it shows the syntactic relationships between words, creates a syntactic structure of the sentence and phrase, as well as the conditions for realizing the lexical meaning of the word. There is usually only one kind of syntax involved subordination. This kind of syntax communicates the relation between the facts of the objective world in the form of a combination of two words, in which one acts as the main and the other as the dependent. Relationships between tokens are represented as lexical-grammatical links between words, which are questions from the main word to the dependent (for example, the operating system). The input to syntax is the results of morphological analysis, presented in the form of multiple pairs li ,i , where li is the token, i is a vector of morphological information token li As a result of parsing, a graph of dependences G is formed at the vertices of which contains tokens. The vertices are joined by arcs that indicate the direction of the link from the parent word to the dependent one.
Statistical text processing is not required for every natural language processing
system. It is usually available in search engines and automated abstracting systems.
These systems are discussed in more detail in the section. Statistical methods are
based on the frequency characteristics of text: the frequency of occurrence of words in
text, the frequency of co-occurrence of several words, the weighted frequency of
occurrence, etc. In these methods, the relations between words are not analyzed
linguistically. During the statistical analysis, the search for input word sequences is
performed and the concepts by which words and phrases are understood, as well as their
frequency characteristics are defined. It is especially important to find the substantive
noun phrases that are given by the schema: consistent word + noun [
The purpose of semantic analysis is to define for each word and phrase some
substantive characterization. The content of the phrase is usually presented as a fragment of
the semantic network. The basis for constructing a fragment is a graph of
dependencies. The result of semantic analysis is the transformation of the dependency graph
into a fragment of the semantic network. The construction of a concept graph
combines two components: the construction of a single semantic network and the
extraction of pragmatic information, that is, the analyzed text extracts its pragmatic content.
Link Grammar Parser (http://www.link.cs.cmu.edu/link) software was used to
construct the conceptual graph of the scientific text. The linguistic support knowledge
base consists of three parts: a fact base, a rule base, and a software knowledge base.
The fact sheet contains a dictionary of ready-made word forms, a dictionary of
endings, a dictionary of inflectional classes and a dictionary of basics. The rule base
consists of the production rules of lexical, morphological, syntactic, statistical and
semantic analyzes. A complete description of all methods is given in [
Ontology describes the concept of particular software and the relationship between
them. In this sense, knowledge becomes possible for re-use by people, databases and
software systems. In addition, the efficiency of both intellectual systems and
traditional information systems is greatly enhanced [
The possibility of automated construction of software ontology is ensured by the acquisition of knowledge of qualitative terminological and / or interpretative dictionaries. In addition, the terminology based on vocabulary knowledge is the nucleus of software ontology. The final version of the ontology should be created with the help of combining several ontology kernels based on different terminology dictionaries. It should be noted that the terminology dictionaries used to create the software ontology should be selected by a knowledge expert. To build complete software ontology, it is necessary to build software nomenclatures that are built on the basis of knowledge from such scientific texts as monographs, textbooks, articles and more. Then you need to combine the terminology and nomenclature. 4.1
There are several classifications (typologies) of dictionaries. The type of any
dictionary is determined by the nature of the lexical material and the practical meaning [
In addition, vocabulary selection vocabularies are distinguished: the dictionaries of the non-vocabulary type and the dictionaries in which the vocabulary is selected according to certain parameters. For example, the scope distinguishes colloquial, voluminous, dialectical, terminological dictionaries. Historically are dictionaries of archaisms, historicisms, neologisms, and more. In terms of disclosure of certain aspects (parameters) words in dictionaries can be - etymological, grammatical, spelling, etc. In terms of revealing the systemic relationships between words, they distinguish nested, word-forming, homonymous, paronymic (expression plan), synonymous, antonymic (content plan) dictionaries. Let us consider dictionaries in terms of their use as a source of knowledge. To build subject ontology requires only qualitative words, containing not only the definition of the term, but also a description of the properties, relations, synonyms and other elements of knowledge about the term. Therefore, it is better to use dictionaries that provide more or less comprehensive scientific information about the subject. Such information is found in encyclopedic, explanatory, terminological dictionaries. Any dictionary consists of dictionary entries. Dictionaries differ in the structure of dictionary articles. Most dictionaries do not have a clear structure for dictionary articles. As a rule, the dictionary article gives one or more definitions (definitions) of concepts, and then describes the relation of the term with other concepts, by which I can explain the essence of the term. These relationships can be generic, part-whole, set a metric for the term, describe the properties of the term, and determine the processes that occur with or over the term something.
The dictionary article of any dictionary starts with the title word, which is the name
of a term or terminological phrase. The title word can be typed in capital letters, bold
or otherwise. The headline is followed by a text that explains the headline in the
dictionary and describes its main characteristics. According to the degree of structure of
the text it is possible to distinguish dictionaries that have a strict structure of
dictionary articles. So a clear structure of the dictionary article has a terminology dictionary
on the basics of computer science and computer engineering [
Interpretation of the sign "concept" t is the centerpiece of the knowledge representation model and is identified with the elemental fragment SF software semantic network: t def , where is semantic network character frame. Since each vertex of such a frame-sign is a vector or a set, it is revealed by a bundle of components of a vector or set, which in turn can also be revealed by a bundle of components. Therefore, a single semantic network of SF character frames will be built. Construction of semantic network of sign-frames, analysis of the constructed network, integration of networks is carried out by means of methods, which are defined in the form of product systems. Initially, products that reveal many of the title terms are
T t activated i . Capitalized terms ti frames of prototype frames of conceptual object "Concepts" are filled, as a result a lot of exoframes is formed Vi | i 1,... T
. So, initially we will have many isolated frames Vi , which, as the slots of the prototype frame are filled innto one network:
SF Vi (3) where n is the number of software terms; Vi is frames that describe all conceptual objects of the software. Therefore, the basic procedure for building a network is a consistent analysis of each glossary of the terminology dictionary, which consists of the following recognition processes: definitions; quantitative relationships; qualitative attitude. 4.3
The following situations arise when mining a definition: 1. A dictionary article may have one or more definitions; 2. If there is one definition in the article, it starts after the first symbol '-', which is encountered in the dictionary article; 3. If there are several definitions in the article, they can be numbered either in Arabic numerals or in the Latin alphabet; 4. In the case of numbering definitions, either the number '.' Or the symbol ')' may be after the number.
Therefore, the beginning of the definition is indicated by the '-' symbol or the appearance symbol "#.", "#)". Here, the '#' symbol indicates an Arabic numeral or Latin letter. In addition, the definition is always found in the first sentence of the dictionary article. This means that to determine the definition, it is necessary to determine whether the first sentence of a dictionary article contains these features.
It should be noted that the direction of research of the automated on (development) structure of ontologies, BR with the help of natural-language texts and systems on their basis is actively developing. In particular, the annual European Conference on Artificial Intelligence holds a meeting of a separate section on ontology training, at which it considers advances in the field of their automated formation. 5
Thus, the analysis of the state of research and development in the field of extraction of knowledge of natural texts is conducted. The general algorithm, the necessary methods and means for extracting new knowledge from the natural text are proposed, it is shown that such algorithm should be multi-stage and include a hierarchical multilevel procedure for recognizing the concepts, connections, predicates and rules that result in ontology from the method performing the recalculation of expected utility.
An ontology development method is built, which is based on the use of an existing ontology in the analysis of text documents used in the construction of the nomenclature and ontology term system. Relationships have been classified and appropriate templates have been developed to search for them in natural language texts. All this made it possible to automate the process of ontology development, which means a significant reduction in costs.