=Paper=
{{Paper
|id=Vol-2386/paper7
|storemode=property
|title=Reflex Intellectual Text Processing Systems: Natural Language Text Addressing
|pdfUrl=https://ceur-ws.org/Vol-2386/paper7.pdf
|volume=Vol-2386
|authors=Serhii Lenkov,Mykola Kubyavka,Liubov Kubiavka,Yevhen Lenkov ,Valerii Shevchuk
|dblpUrl=https://dblp.org/rec/conf/momlet/LenkovKKLS19
}}
==Reflex Intellectual Text Processing Systems: Natural Language Text Addressing==
https://ceur-ws.org/Vol-2386/paper7.pdf
Reflex Intellectual Text Processing Systems: Natural
Language Text Addressing
Serhii Lenkov [0000-0001-7689-239X], Mykola Kubyavka [0000-0003-1745-7026],
Liubov Kubiavka [0000-0002-5141-9886], Yevhen Lenkov [0000-0002-3784-5790],
Valerii Shevchuk [0000-0002-6950-4827]
Military Institute of Kyiv National Taras Shevchenko University, Ukraine
Kyiv National Taras Shevchenko University, Ukraine
Central Research Institute of the Armed Forces, Ukraine
University of the State Fiscal Service, Ukraine
lenkov_s@ukr.net, nick.kub@ukr.net, klb_it@ukr.net,
torwer007@gmail.com, shevchuk764@ex.ua
Abstract . The article describes the method of automatic response to the
content of the text of the message, which was based on a probabilistic-reflexive
approach. The reflexive approach provided the choice of the most probable
response to the set of input influences, with known probabilities of choosing
the response for each input effect, as well as some combinations of input
influences, and the method developed on its basis allowed to automatically
determine the destination of the analyzed text.
Keywords : Information Technology; natural language text; reflexive
approach; methods of information management.
1 Introduction
The ability to think is a true wealth, only the presence of strong feelings and thoughts
proves you do not stand still, these thoughts are helping us to develop more than time
and space, against those we are nothing. Thoughts, views and dreams are changing as
we speak. It depends on our surrounding, mood and emotions. Leo Tolstoy said: "Any
thought, expressed in words, is a force whose action is boundless" [1].
Our consciousness may not be limited to words, but we are able to convey
thoughts, emotions and wishes to people taking in the form of words. Language
captures our knowledge of the world and allows us to express our attitude towards it. It
is the main means of human intercourse. Key words or simple patterns will not be able
to cover the needs of modern man on natural language text information processing. As
has been held by several scholars, a solution could be attributed to the use of artificial
intelligence. Thus, the information extraction (meaning of the text) or reading
comprehension are one of the challenging tasks in the field of information technology
and artificial intelligence. Although there are plenty studies on the construction of
natural language processing systems [2-14], which arose by merging of computer
linguistics and information retrieval systems [2]. Nevertheless, the implementation of
problem-oriented intellectual systems that would be capable of selecting the necessary
information from the natural language and applying this correctly was and remains a
relevant problem of the present.
2 Research results analysis
Any information we undergo shape our entire lives and that could be expressed as text.
This paper discusses the natural language texts processing and analysis using the
influence of such information to us instead of basic lexicon-semantic approach.
Different information affects differently, and this is due to a phenomenon such as its
addressness. If the message is not correctly addressed it will not be understandable,
and on the contrary, the address-oriented message will affect the recipient and will
allow him to understand its meaning.
The information management models and methods [15] based on the results of the
interaction theory in the context of means of processing natural linguistic information
will be used to address the problem.
The reflex probabilistic approach was used to identify the most probable reaction to
the incoming message, with known probabilities of choosing the reaction for each
input effect, as well as some combinations of input effects [12]. On its basis, the reflex
method of automatic response to the message content and the model for determining
the most informative components of the natural language text are developed. Any
natural language text is considered as a message perceived by the intellectual
apparatus of a person.
Here are some definitions to take the point further.
Definition 1. Text addressness - is the semantic component of the information, which
determines the information direction of the message content, has corrective influence
on the awareness (world view) of the recipient.
Definition 2. The message recipient can be determined from the message itself.
Definition 3. The concept of text length is a quantitative indicator of information and
conceptual segments in the message part (text).
Definition 4. According to the information disseminated (contractors), it should be
understood that the notice is communicated and does not lead to changes in these
consumers.
By notice, it will mean any information (telecast, newspaper articles, internet notes,
etc.) that will be sent to consumers. It`s received receptions (can take) necessary
decisions for us. For the entire flow of information (all messages) used and
technologies that provide impact on the interaction contractors, it is necessary to
distribute to the address. It has also been recognized as the classical interaction
partners (consumer information), and which classical messages will be published in
the names of the technologies. Please inform that the information about who to send
the message is in the message itself.
Definition 5. Identification of the necessary relation to the received information - is
the definition of the actual readiness of the recipient to take one or another decision in
a given situation in the direction of the desired party, exercising influence.
Definition 6. Identification of forms, methods and content of information influence on
counteragents of interaction - is determining the ways of providing information and the
content of the message itself, which will provide targeted impact on the target
audience in accordance with the objectives of influence. The recipient must also
provide false information if he so affects him that he will take the necessary
(appropriate for the purpose of the effect) decision.
Definition 7. Managing impulse awareness - is organizing, planning and verifying
such information actions for the recipient who can form an adequate target for the
impact of the relation to the actual and predicted situations. This, in turn, will increase
the likelihood of making the right decisions.
And so, the first step, it is necessary to decompose the input text into clusters,
among which there will usually be those that determine the necessary reaction under
the given conditions. Since it is impossible to know in advance what these clusters are,
we will create a series of combinatorial regularities of the input text and select the ones
that most affect the recipient.
The acknowledgement that the informational influence of the natural language text
leads to an increase or decrease in the probability of the desired reaction in some
classes of recipients is the basis of the paper:
pk ( Al ) pk ( Al / I j ),
where A i is the recipient reaction (behaviour); I j is information influence of the
message; Pk (A i ) is absolute probability of the reaction (behaviour) A i in the class of
recipients Q k, if this class does not get influenced; p k (A/ lj) i s the reaction
(behaviour) probability A l in the class of recipients Q k after the influence lj.
In addition to the determination of its impact on the recipient, it is necessary to
select for each message such classes of recipients, which impact will be maximal. For
the given lj and A l we should choose a class of recipients Q*, which impact p k (A /l j)
will be maximal:
Q * max pk ( Al / I j ) ,
k
where Q * i s the class of recipients is chosen to implement the impact.
Thus, the function finding that determines the difference between the reaction
probability R t for the recipient class Q ' without influence and with that Aj is a high-
priority problem.
The information influences repetition is a way to solve this problem, but in practice
there had been no such cases. Each effect is combinatorial and usually consists of
many separate, sometimes insignificant elements. Then this problem will be solved in
the following way. If message is informative and significantly increases the reaction
probability, it means that the message parts (separate sentences, words, combination of
letters) are the influence holders on the counterparty. When these parts arise, the
reaction probability changes. However, as stated above, the messages are not actually
repeated, but the parts definitely are. Therefore, the probability change of one or
another reaction can be assessed not from the whole message, but from some of its
part.
To obtain this information, the following method is used.
1. An expert is provided with a text message and a list of recipient classes with their
properties.
2. The expert assesses the information and provides for such messages the most likely
addressees (whose probability of the desired reaction will be changed).
3. The statistics accumulation is made possible through the involvement of many
experts.
4. Based on the information received, recipients for new text messages are found.
A solution to the problem 4 requires the development of method by which the
addressees for new messages are determined on the basis of an expert assessment.
Existing methods of solving this problem are quite complex, require the use of
linguistic analysis tools that are almost impossible to use for messages of different
directions, or their creation will be too expensive, since linguistic systems are usually
created for separate text classes.
One of the varieties of intellectual systems is reflex intellectual systems, which are
nothing more than software or technical systems that form reactions to non-power
influences, the basis of which is the reflex algorithm that operates on the principle of
forming reaction-response (reflex) on the set input data (external influence). But this
approach has not yet been used by anyone to identify the addressees of the message.
Therefore, it is proposed to use the probabilistic-reflexive approach to solve the
problem. The basis of the reflexes is the following thesis: if it already was, and some
reaction was positively reinforced, then it is necessary to do the same.
The reflexion approach provides the choice of the most probable reaction to the
infinity of input influences, with known probabilities of choosing the reaction for each
input effect, as well as some combinations of input influences.
Considering the method of counterparty selection, where the probability of the
desired reaction is greater than the limit value. The method is based not on the
probability calculation of addressing messages, but on the assessment of this
probability for the deviation of conditional and unconditional probabilities of such
addressing from parts of natural language text. The use of conceptual information and
relevant content forms in the method as well as means of information influence are
complementary, not alternative. This would affect the method of selecting the
necessary objective-oriented information is used to calculate the amount of
information action on the recipient.
Definition 7. The method of determining the message recipients - is a search, selection
and presentation method of relevant information to a target audience, the application of
which will lead to the necessary responses to those changes the awareness (world
view) of the target audience.
The addressed information is the result produced by the intellectual apparatus of a
person. The semantic component of information, which is the core and product of
communication in the intellectual apparatus of a person, determines its addressness.
The correct definition of the natural language text addressness is response to the
message. The addressness implies that one should understand the person to whom this
message is addressed.
Definition 8. The addressed information is the result produced by the intellectual
apparatus of a person. The semantic component of information, which is the core and
product of communication in the intellectual apparatus of a person, determines its
addressness. The correct definition of the natural language text addressness is response
to the message. The addressness implies that one should understand the person to
whom this message is addressed.
The reaction caused in the process of the addressness determining may be wrong.
And furthermore, the same result (assess of one or another addressness) may
correspond to the reality or be disconnected with that at all.
When using the probabilistic-reflexive approach, it is necessary to determine what
is the influence, and what are the reflexes in the means of information technology.
From definition 2, the influence is the natural language text (message), and from
definition 4 it follows that the reaction is a class of recipients. Then the possibility of
obtaining the correct addressee can be represented through a certain probability.
pi bi / ki , (1)
where p t i s the probability that the produced will necessarily bring the desired result
(correctly addressed message); bi is tests conducted to correctly identify the
information addressness; ki i s total number of tests.
Determining the addressness of text messages based on the evaluation of the results
of past actions can be presented in the form below.
Table 1. Statistical data of the response development to determine the
text addressee
The number of times the
The number of times the
Addressness targeting of text clusters was
desired result was obtained
determined
A1 k1 b1
A2 k2 b2
… … …
Ai ki bi
… … …
Am km bm
First, you need to determine what affects the addressness.
Definitions 1 and 2 provide an answer to this question. The message along with the
natural language text fragments, by which that could be transmitted. Such fragments
are called the elementary influence unit.
Definition 9. The elementary influence unit will be considered a separate symbol,
letter combination, composition, word, phrase, sentence, text excerpts, image,
statement, sound, etc., that smallest piece of information provided to the recipient for
perception is perceived by him and leads to a complete change of his world view.
Through the fragments in messages addressed to one or another class of recipients,
the probability of sending a message to these recipients can be estimated.
Numerical measure of influence on the characteristics of addressness text
determining is used as a basis for solving the problem. The examination of various
influences on the process of the information addressing identification will help to
identify the dependence of the automatically determined information addressness from
the addressnes determined by an expert.
An existing model for identifying the most informative components of the natural
language text will be applied to assess the probability of a well-defined recipient.
The message is decomposed into numerous fragments:
I i , f 1, g , (2)
f
where I is information (message); if i s the often-repeated fragment; g i s the number
of fragments.
Successfully identifying the necessary information addressing will facilitate the
determination of approximate probability values of each of the addressing information
classes subject to the availability of new addressing fragment.
A j A, i I : p ( A j / i ) n ( A j / i ), (3)
f f f
where n ( A j / i ) i s the frequency of the addressness determining A g subject to the
f
availability of the fragment; p ( A j / i ) i s the addressnes determining probability A g
f
subject to the availability of the fragment if.
By addressing Ai under its unconditional probability p (A j) and partial conditional
probabilities p ( A j / i ),..., p ( A j / i ),..., p ( A j / i ) the compatible conditional
f f f
1 i g
probability can be estimated:
p( A j / i ,..., i ,..., i ) p ( A j / I ), (4)
f f f
1 i g
The problem of finding a compatible conditional probability of a correctly
determined information addressing cannot be solved by using the probability theory
methods, taking into account the partial conditional and unconditional probabilities.
The first step is to define the issue clearly in order to estimate a compatible conditional
probability by partial. We should apply the evaluation method for compatible
conditional probability by partial that will enable us to select the addressing that would
have been determined by the expert. For instance:
Ag AAg p ( Мо ( Ag / I ) Мо ( Ag / I ) / p ( Ag / I ) p ( Ag / I )) 1, (5)
where ηMo(Ag / I) i s assessment of the compatible conditional addressing probability
A g of information I , which was obtained by the evaluation method for compatible
conditional probability by partial; Mo is the evaluation method for compatible
conditional probability by partial; p ( Мо ( Ag / I ) Мо ( Ag / I ) / p ( Ag / I ) p ( Ag / I )) i s the
conditional probability whereby conditional probability of information addressnes
determining A g reaches maximum, then the compatible conditional probability
evaluation of this addressing A g is maximal.
Definition (7) shows that if the compatible conditional probability of the recipient
determining Ag reaches maximum, then almost always its evaluation is maximal. An
optimal Mo method always gives the highest evaluation of the greatest p(Ag / I),
p ( A j / i ),..., p ( A j / i ),..., p ( A j / i ) , can be obtained compatible conditional
f f f
1 i g
probability. To assess the effectiveness of this method the experimentally.
The deviation of this equation from the unit is a criterion for the effectiveness of
the evaluation method for compatible conditional probabilities by partial:
p ( Мо ( Ag / I ) Мо ( Ag / I ) / p ( Ag / I ) p ( Ag / I )) max, , with restrictions:
• the data collection I is decomposed into numerous fragments I f ,
• there is information with the highest evaluation of addressing information influence
Ai;
Ag A : p ( Ag / i ),..., p ( Ag / i ),..., p ( Ag / i );
f f f
1 i g
A g AA g A : p ( A g / I ) p ( A g / I );
Mo
Eg A : p ( Ag / i ),..., p ( Ag / i ),..., p ( Ag / i )
Мо ( Ag / I ); (6)
f f f
1 i g
A mathematical model presented in the paper [16] will be applied not in relation to
influence on the counterparty, but to the addressee information identifying.
Using values from Table 1 in the formula (1) of work [16] we obtain a new formula
for assessing the influence of some fragment if on the message addressing Ai:
bi 1 p ( Ai )
1. If ki bi bi 0 : w( Ai / i f )
; (7)
k i bi p ( Ai )
(bi 1) 1 p ( Ai )
2. If k i bi bi 0 : w( Ai / i f )
k i bi 1 p( Ai )
, (8)
where w( Ai / i f ) i s influence assessment of the message fragment i f on the
response (the addressee choice) Ai; bi i s number of tests, when the necessary
information addressing was correctly determined; ki is total number of tests.
Every message consists of numerous fragments. Then the total effect will be
determined from the formula (1) [7]:
w( Ai / I )
g
w( A / i ) 1 2 g
w( Ai / i f ) 1
f 1
i f
w( Ai / i f )
f 1 w( Ai / i f )
, (9)
where w( Ai / I ) i s evaluation all influences on the addressee choice.
Using these formulas, estimation of the probability of choice we can proceed the
assessment of the new probability of addressee choice Ai:
w( Ai / I ) p ( Ai )
p ( Ai / I ) , (10)
1 p ( Ai ) ( w( Ai / I ) 1)
where p ( Ai / I ) i s the evaluation probability of the addressee choice Ai.
Since there are many recipients, the probability is estimated for each of them. The
recipients whose estimated probability is more than 0.5 are chosen.
The expert information is needed on how the message classes are addressed to the
recipients (recipient classes). Let's consider this question.
To determine the quantitative measure of the message classes (forms) influence on
recipient classes, expert analysis of processes accompanied decisionmaking on the
message implementation in various influence components and the various functional
roles performance has been undertaken. Experts were asked to identify the subjective
probability of selecting one or another form of text messages in various influence
components and the functional roles implementation.
Definition 10. Subjective probability - is an expert assessment of the probability of a
particular event occurrence.
A sociological survey of school-aged children, preschool-age children and
psychologists working in these educational institutions was conducted, an analysis of
social surveys on the Internet and the media on these issues, a survey of students and
elderly, as well as experts in the field of sociological research, the integrated expert
assessment of the subjective probability of implementing one of the model attributes
was formulated. The subjective expert relation to the implementation of various
message forms in different conditions could be considered in the context of
probabilities.
In the process of analysis, subjective-probabilistic performance indicators were
identified:
• the class of messages for children;
• the class of messages for working youth;
• the class of messages for young people;
• the class of messages for entrepreneurs;
• the class of messages for employees;
• the class of messages for servicemen;
• the class of messages for pensioners.
The subjective-probabilistic implementation indicators of the message class were
found to be applicable to recipient classes [16], and the data obtained were included in
the expert tables.
A recent analysis is a basis for assessing the information actions importance in the
process of influence management. This contributed to identifying those forms of text
messages that can maximally effectively ensure the implementation of these actions on
the recipients. The information provided in the expert tables has become the basis for
determining the necessary in formation of optimal, rational or appropriate
informational influences on recipients by means of communication. The effective
implementation of message tools requires the methods development for identifying
information actions and the required forms of communication based on the received
subjective-probabilistic characteristics of the scope of application.
The expected outcome of this task is a summary table, in a situation or if there is a
problem in the influence management process, evaluating one of the influence
components and knowing the information type (functional role), the most suitable and
informative message (natural language text) can be easily identified. To find a text that
will form the necessary relation to the reality (necessary knowledge) of the recipient
would be easier with those summary tables. The use of the above method helps to
identify the recipient and the text defining to be sent to the specified addressee.
This method is not about "understanding" of a limited natural language, but system
that perceives information representations in a natural language without restrictions.
3 Conclusion
In summary, the authors consider it is necessary to mention such remarkable features
of human intelligence as its extraordinary flexibility and mobility. Indeed, as soon as a
person learns something, the field of his knowledge immediately expands. One of the
main aspects in human learning is not the more he learns the better he can solve a new
task, but rather to be able to match, join and combine with all the previous tasks in
order to make a decision. The so-called reflex intellectual systems are based on this
peculiarity of human consciousness, which are nothing more than software or technical
systems that form reactions to influences, based on the reflex algorithm built on
introformational methods [12] and works on the principle of forming reaction-response
(reflex) on the set of input data (external influence).
The mathematical apparatus of the interaction theory [12] is quite simple and
convenient for the implementation of processing natural language text systems. During
the implementation of the above method, existing models for identifying the most
informative components of the natural language text from the position of automatic
addressing of these messages to different classes of recipients have been improved.
The model innovativeness is to use a probabilistic-reflex approach to determine the
natural-language text addressee. The proposed model differs from the existing in the
difference identification of the conditional probabilities of the recipients occurrence as
the information influence extent of the natural language text fragments, which
automatically identified the most likely addressees of this text.
The developed technology meets the requirements for simplicity, is informatively
clear and minimal costly. Its foundation is based on the models and methods of
influence management [15, 16], and the results are getting closer and closer to
mankind to create a complete artificial intelligence.
4 References
1. Templeton J.: World laws of life, http://www.mirradosti.com.ua/?p=1669, last access
2019/05/27
2. Popov E.: Communication with a computer in a natural language. Nauka, Moskva (1982).
3. Bolshakov E., Lande D., Noskov A., Klyshynskyy E., Peskova A., Yahunova E.: Naturally
texts automatic processing on language and computer linguistic. MYEM Publ, Moskva
(2011).
4. Brill E., Mooney R. J.: An Overview of Empirical Natural Language Processing. In: AI
magazine, vol. 18(4), pp. 13-24 (1997).
5. Pasichnik V., Vysotska V., Shestakevych T., Shcherbina Y.: Mathematical linguistics. New
World Publ., Lvov, Ukraine (2012).
6. Fedorchuk A.: Content Monitoring of information flows. In: Nat. Acad. Science Problems.
Functioning, Trends of development, vol. 3(12) (2005).
7. Grishman R.: Computational Linguistics: An Introduction. Cambridge etc.: Cambridge
University Press (1986).
8. Bondarenko M.: Osnovy teoriyi syntezu nadshvydkodiyuchykh struktur movnykh system
shtuchnoho intelektu. Kiev, Ukraine (1997).
9. Belonogov G.: Komp'yuternaya lingvistika i perspektivnyye informatsionnyye tekhnologii.
In: Nauchno-tekhnicheskaya informatsiya, vol. 2(8), pp. 30-43 (2004).
10. Teslya Yu., Kosenkova O.: Zastosuvannya teoriyi informatsiynoyi vzayemodiyi do
pobudovy system smyslovoho poshuku informatsiyi. In: Radioelektronika i informatika.
vol. 1, pp.103-105 (2000).
11. Fayn V.: Raspoznavaniye obrazov i mashinnoye ponimaniye yestestvennogo yazika. Nauka,
Moskva (1987).
12. Teslya Yu.: Vvedeniye v informatiku prirody. Maklaut, Kiev, Ukraine (2010).
13. Teslya Yu., Timchenko A.: Opyt razrabotki i primeneniya v stroitel'stve instrumental'nykh
programmnykh sredstv yestestvenno-yazykovogo obshcheniya. In: Tezisy dokladov na II
mezhdunarodnom nauchno-tekhnicheskom seminare "Teoreticheskiye i prikladnyye
problemy modelirovaniya predmetnykh oblastey sistemakh baz dannykh i znaniy". MGP
"Tirazh". Tuapse, pp. 226-228 (1993).
14. Cunningham H., Humphreys K., Gaizauskas R., Wilks Y.: Software Infrastructure for
Natural Language Processing. In: Proc. of the 5th Conference on Applied Natural
Language Processing. Washington, pp. 237-244 (1997).
15. Kubiavka M.: Modeli ta metody upravlinnya informatsiynym suprovodzhennyam v
umovakh hibrydnoyi viyny: synopsis for Ph.D. thesis in Engineering Science: 05.13.06,
Kyiv. Taras Shevchenko nat. uni. of Kyiv (2017).
16. Kubiavka M.: The control of information support of the processes of influence on the
opponent. Zbirnyk naukovykh prats Viiskovoho instytutu Kyivskoho nats. Univers. imeni
Tarasa Shevchenka. vol. 54, pp. 174-181 (2017).