An intelligent chatbot for evaluating the emotional colouring of a message and responding accordingly
A recurrent neural network model, a database designed for neural network training, and a software tool for interacting with a bot have all been created. A large dataset (50 thousand comments) containing different reviews and sentiments was collected and annotated to successfully train and validate t...
Збережено в:
| Дата: | 2024 |
|---|---|
| Автори: | , , , , , |
| Формат: | Стаття |
| Мова: | English |
| Опубліковано: |
Інститут програмних систем НАН України
2024
|
| Теми: | |
| Онлайн доступ: | https://pp.isofts.kiev.ua/index.php/ojs1/article/view/603 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Problems in programming |
| Завантажити файл: |  |
Репозитарії
Problems in programming| id |
pp_isofts_kiev_ua-article-603 |
|---|---|
| record_format |
ojs |
| resource_txt_mv |
ppisoftskievua/bd/7d79ef3332c78ef9efdeb04b4a4918bd.pdf |
| spelling |
pp_isofts_kiev_ua-article-6032024-04-27T17:01:04Z An intelligent chatbot for evaluating the emotional colouring of a message and responding accordingly Розумний чат-бот для оцінки емоційного забарвлення повідомлення та відповідної відповіді Kobchenko, V. R. Shymkovysh, V.M. Kravets, P.I. Novatskyi, A.O. Shymkovysh, L.L. Doroshenko, А.Yu. recurrent neural networks; sentiment analysis; Python; tensorflow; keras; chatbot UDC 004.8 рекурентні нейронні мережі; аналіз настроїв; Python; tensorflow; keras; chatbot УДК 004.8 A recurrent neural network model, a database designed for neural network training, and a software tool for interacting with a bot have all been created. A large dataset (50 thousand comments) containing different reviews and sentiments was collected and annotated to successfully train and validate the model. It was also translated into Ukrainian language with the help of an automatic translator. The architecture of the neural network model underwent optimization to enhance classification outcomes. Furthermore, work was conducted on enhancing the user interface. The developed application was tested, and the results were demonstrated. The resulting model demonstrated accuracy 85% in determining sentiments. The implemented application has got basic design (which can be customized) and some settings for chatbot. Further improvement of the model’s classification quality can be achieved by collecting a larger and better organised dataset or by researching other RNN architectures.Problems in programming 2024; 1: 23-29 Створено рекурентну модель нейронної мережі, базу даних, призначену для навчання нейронної мережі, програмний інструмент, що її реалізує, для взаємодії з ботом. Для успішного навчання та перевірки моделі було зібрано та анотовано великий набір даних, 50 тисяч коментарів, що містять різні відгуки та думки. Його перекладено українською мовою за допомогою автоматичного перекладача. Архітектура моделі нейронної мережі була оптимізована для покращення результатів класифікації. Крім того, була проведена робота над вдосконаленням інтерфейсу користувача. Розроблений додаток протестовано, результати продемонстровано. Отримана модель продемонструвала точність 85% у визначенні настроїв. Реалізована програма має базовий дизайн, який можна налаштувати, а також деякі налаштування для чат-бота. Подальшого покращення якості класифікації моделі можна досягти шляхом збору більшого та краще організованого набору даних або шляхом дослідження інших архітектур RNN.Problems in programming 2024; 1: 23-29 Інститут програмних систем НАН України 2024-04-01 Article Article application/pdf https://pp.isofts.kiev.ua/index.php/ojs1/article/view/603 10.15407/pp2024.01.023 PROBLEMS IN PROGRAMMING; No 1 (2024); 23-29 ПРОБЛЕМЫ ПРОГРАММИРОВАНИЯ; No 1 (2024); 23-29 ПРОБЛЕМИ ПРОГРАМУВАННЯ; No 1 (2024); 23-29 1727-4907 10.15407/pp2024.01 en https://pp.isofts.kiev.ua/index.php/ojs1/article/view/603/665 Copyright (c) 2024 PROBLEMS IN PROGRAMMING |
| institution |
Problems in programming |
| baseUrl_str |
https://pp.isofts.kiev.ua/index.php/ojs1/oai |
| datestamp_date |
2024-04-27T17:01:04Z |
| collection |
OJS |
| language |
English |
| topic |
recurrent neural networks sentiment analysis Python tensorflow keras chatbot UDC 004.8 |
| spellingShingle |
recurrent neural networks sentiment analysis Python tensorflow keras chatbot UDC 004.8 Kobchenko, V. R. Shymkovysh, V.M. Kravets, P.I. Novatskyi, A.O. Shymkovysh, L.L. Doroshenko, А.Yu. An intelligent chatbot for evaluating the emotional colouring of a message and responding accordingly |
| topic_facet |
recurrent neural networks sentiment analysis Python tensorflow keras chatbot UDC 004.8 рекурентні нейронні мережі аналіз настроїв Python tensorflow keras chatbot УДК 004.8 |
| format |
Article |
| author |
Kobchenko, V. R. Shymkovysh, V.M. Kravets, P.I. Novatskyi, A.O. Shymkovysh, L.L. Doroshenko, А.Yu. |
| author_facet |
Kobchenko, V. R. Shymkovysh, V.M. Kravets, P.I. Novatskyi, A.O. Shymkovysh, L.L. Doroshenko, А.Yu. |
| author_sort |
Kobchenko, V. R. |
| title |
An intelligent chatbot for evaluating the emotional colouring of a message and responding accordingly |
| title_short |
An intelligent chatbot for evaluating the emotional colouring of a message and responding accordingly |
| title_full |
An intelligent chatbot for evaluating the emotional colouring of a message and responding accordingly |
| title_fullStr |
An intelligent chatbot for evaluating the emotional colouring of a message and responding accordingly |
| title_full_unstemmed |
An intelligent chatbot for evaluating the emotional colouring of a message and responding accordingly |
| title_sort |
intelligent chatbot for evaluating the emotional colouring of a message and responding accordingly |
| title_alt |
Розумний чат-бот для оцінки емоційного забарвлення повідомлення та відповідної відповіді |
| description |
A recurrent neural network model, a database designed for neural network training, and a software tool for interacting with a bot have all been created. A large dataset (50 thousand comments) containing different reviews and sentiments was collected and annotated to successfully train and validate the model. It was also translated into Ukrainian language with the help of an automatic translator. The architecture of the neural network model underwent optimization to enhance classification outcomes. Furthermore, work was conducted on enhancing the user interface. The developed application was tested, and the results were demonstrated. The resulting model demonstrated accuracy 85% in determining sentiments. The implemented application has got basic design (which can be customized) and some settings for chatbot. Further improvement of the model’s classification quality can be achieved by collecting a larger and better organised dataset or by researching other RNN architectures.Problems in programming 2024; 1: 23-29 |
| publisher |
Інститут програмних систем НАН України |
| publishDate |
2024 |
| url |
https://pp.isofts.kiev.ua/index.php/ojs1/article/view/603 |
| work_keys_str_mv |
AT kobchenkovr anintelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly AT shymkovyshvm anintelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly AT kravetspi anintelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly AT novatskyiao anintelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly AT shymkovyshll anintelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly AT doroshenkoayu anintelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly AT kobchenkovr rozumnijčatbotdlâocínkiemocíjnogozabarvlennâpovídomlennâtavídpovídnoívídpovídí AT shymkovyshvm rozumnijčatbotdlâocínkiemocíjnogozabarvlennâpovídomlennâtavídpovídnoívídpovídí AT kravetspi rozumnijčatbotdlâocínkiemocíjnogozabarvlennâpovídomlennâtavídpovídnoívídpovídí AT novatskyiao rozumnijčatbotdlâocínkiemocíjnogozabarvlennâpovídomlennâtavídpovídnoívídpovídí AT shymkovyshll rozumnijčatbotdlâocínkiemocíjnogozabarvlennâpovídomlennâtavídpovídnoívídpovídí AT doroshenkoayu rozumnijčatbotdlâocínkiemocíjnogozabarvlennâpovídomlennâtavídpovídnoívídpovídí AT kobchenkovr intelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly AT shymkovyshvm intelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly AT kravetspi intelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly AT novatskyiao intelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly AT shymkovyshll intelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly AT doroshenkoayu intelligentchatbotforevaluatingtheemotionalcolouringofamessageandrespondingaccordingly |
| first_indexed |
2024-09-16T04:08:51Z |
| last_indexed |
2024-09-16T04:08:51Z |
| _version_ |
1818568518749126656 |
| fulltext |
Експертні та інтелектуальні інформаційні системи
23
© V. R. Kobchenko, V.M. Shymkovysh, P.I. Kravets, A.O. Novatskyi, L.L. Shymkovysh, А.Yu. Doroshenko, 2024
ISSN 1727-4907. Проблеми програмування. 2024. №1
UDC 004.8 http://doi.org/10.15407/pp2024.01.23
V. R. Kobchenko, V.M. Shymkovysh, P.I. Kravets, A.O. Novatskyi, L.L. Shymkovysh,
А.Yu. Doroshenko
AN INTELLIGENT CHATBOT FOR EVALUATING
THE EMOTIONAL COLOURING OF A MESSAGE
AND RESPONDING ACCORDINGLY
In this article, a recurrent neural network model, a database designed for neural network training, and a
software tool for interacting with a bot have all been created. The architecture of the neural network model
underwent optimization to enhance classification outcomes. Furthermore, work was conducted on enhancing
the user interface. The developed application was tested, and the results were demonstrated. The resulting
model demonstrated 85% accuracy in determining sentiments. The implemented application has got basic
design which can be cutomized and some settings for chatbot. Further improvement of the model’s
classification quality can be achieved by collecting a larger and better organised dataset or by researching other
RNN architectures.
Key words: recurrent neural networks, sentiment analysis, Python, tensorflow, keras, chatbot.
Introduction
Emotions in the broad sense are
psychological and physiological reactions that
arise in a person due to certain events, stimuli
or thoughts. They are reflected through
internal experiences and external expressions.
Usually, it is very easy to determine a
person's emotions by observing their
behavior, looking at their faces, or listening to
their voices. However, the text itself does not
have an emotional color. Instead, it can evoke
emotions in the reader and reflect or express
the emotional state of the author. This makes
the task of assessing emotionality more
difficult. And we usually don’t need to know
exact emotions, it is enough to know are they
positive or negative.
This task is known as sentiment analysis
[1,2] and is used in many areas of life. For
example, processing reviews for products in a
store can greatly simplify the work of
marketers.
Besides marketing it is used in psychology,
both for usual conversations and professional
online help by psychologists.
And while it is not hard task for people, it
can be still useful to automatize such analysis
[3]. For marketers it will help, because they
have to work with big number of reviews.
Automatic detection of sentiments can facilitate
the work of psychologists too. And this is
possible to do with neural networks [4,5].
Artificial neural networks (ANNs) are
computer models that imitate the functioning
of biological neural networks [6,7].
Structurally they consist of interconnected
artificial neurons that exchange signals in the
form of numbers. There can be hundreds,
thousands or even millions of such neurons.
Each neuron receives input data, performs a
certain function on this data and transmits the
result to the next neurons in the network.
ANNs can be used in a wide range of
applications, such as natural language
processing, computer vision and speech
recognition [7-17].
Recurrent Neural Networks (RNNs) are a
special class of artificial neural networks
[18,19]. Their feature is that they take into
account the contextual dependence between
elements of the sequence, that is, information
about previous states to be taken into account
in the current state.
In recurrent networks a directed sequence
of elements is formed, presented in the form
of a directed graph, and neurons are able to
transmit it together with the processed data.
This is achieved by using loops in each
neuron.
Експертні та інтелектуальні інформаційні системи
24
1. Development of RNN
As already mentioned, the main feature of
recurrent networks is the possibility of using
pre-processed information for a better
understanding of the existing information.
This is a big step forward and it is actually
effective. But does it always work? It all
depends on how far down the sequence the
context we need now is. If this "distance" is
small, then everything really works well.
But it is quite clear that this is usually not
the case and this "distance" can often be
very large. At the same time, with its
increase, the network at some point
becomes unable to learn and combine the
necessary information.
This problem is known as the gradient
vanishing problem and is one of the main
problems of recurrent neural networks.
Various modifications have been developed
to overcome this problem. LSTM (Long Short-
Term Memory) is one of those [20].
LSTM is an improvement of the recurrent
neural network architecture that allows to
study of long-term dependencies [21-23].
This allows such networks to cope with a
large number of problems, thanks to which
they have been widely used.
All recurrent neural networks have the
form of a chain of repeated neural network
modules. In standard RNNs, this chain has a
simple structure, usually one layer with a
certain activation function: hyperbolic
tangent, sigmoid, etc.
LSTM also has a chain-like structure, but
the module structure is different. Instead of
one layer, there are four layers that interact
with each other as shown in the figure 1.
Figure 1. A recurrent LSTM module with four layers [24]
In the figure 1 each line carries a vector
from the output of one node to the inputs of
the others [24]. Pink circles represent
element-by-element operations such as vector
addition, and yellow rectangles represent
trained neural network layers. Merged lines
indicate concatenation, while split lines
indicate copying of content, which is then
sent to different locations.
The main idea behind LSTM is the cell
state, the horizontal line that runs along the
top of the chart.
The state of the cell is like a conveyor belt.
It is simply laid along the entire chain with
minor linear interactions. Information easily
passes along the entire tape without changes.
But LSTMs have the ability to remove or
add information to the state of a cell, which is
controlled by structures called gates.
Filters are a way of controlling the
transmission of information. They consist of a
sigmoidal neural network layer and an
element-by-element multiplication operation
between the results obtained from the
sigmoidal layer and the current state of the
cell state.
The sigmoid layer outputs a number
between zero and one, describing how much
of each component should be transmitted. A
Експертні та інтелектуальні інформаційні системи
25
value of zero means "skip nothing" and a
value of one means "skip everything".
LSTMs have three such filters to control
the state of the cell.
The cell value is calculated in several steps.
First, analyzing the information from the
previous cell and deciding what can be removed
from it. This decision is made by a sigmoidal
layer called the "forgetting filter layer".
The next step is to decide what new
information we will store in the cell state, and
it consists of two parts. Firstly, the sigmoid
layer, which called the "input filter layer",
decides which values need to be updated.
Secondly, the hyperbolic tangent (tanh) layer
creates a vector of new values that can be
added to the state.
After that updating the old state of cell Ct-1
and obtaining a new Ct are performed. To do
this, we multiply the old state by the value of
ft (the "forgetting filter layer"), thus forgetting
what was decided to be forgotten in the first
step. Then we add the product of it and Ct.
These are the new potential values multiplied
by the "importance" factor, which is how
much we decide to update each state value.
Finally, we need to decide what we're
going to output. This will be based on our cell
state, but should also be filtered. First, a
sigmoid layer is applied, which decides which
parts of the cell's state we pass next. Then we
pass the cell state through the tanh layer so
that the values are in the range of -1 to 1 and
multiply by the output values of the other
sigmoid layer to output only what is needed.
This is the basic idea of LSTM. These have
the ability to remember long-term dependen-
cies, delete and add information to the cell's
state, and control what will go to the output.
2. Development of a dataset for
the training of model
This section describes the data collection
process for training the LTSM model.
Data collection and preparation is an
important stage in the process of developing
and training a neural network [25], because
we need to prepare the data on which it will
learn.
The online platform Kaggle can help with
this. Here you can find a large collection of
various datasets for any purpose. There is also
a proprietary development environment that
can also be used to build neural networks. At
the same time, the platform unites a
community of people engaged in machine
learning. Here you can ask anything on this
topic or discuss something. Various
competitions are also often held.
A dataset named IMDB Dataset of 50K
Movie Reviews was found in Kaggle [26].
This is a collection of 50,000 movie reviews,
each of which is marked with a sentiment
(positive or negative) as shown in the figure 2.
The dataset is well balanced, so we have an
almost equal number of positive and negative
reviews.
Since the bot, which is being developed,
must communicate in Ukrainian, the dataset
was translated using the googletrans library
for Python, which provides access to Google
Translate from a python code file. Of course,
the translation cannot be called exact, but
Figure 2. Content of the dataset
Експертні та інтелектуальні інформаційні системи
26
translating 50 thousand reviews by hand
would take too much time, so it was used.
The data also includes a list of stop words
in the Ukrainian language. Fortunately, there
are already ready-made lists, so one of these
was used.
In addition, files were prepared with certain
variants of appeals to the bot and probable
reactions to such appeals. Scripts include
greetings, farewells, thanks, and questions.
And separately, of course, there is all the other
text, which can be positive, negative or neutral,
depending on which the bot will have a
different reaction.
3. Description of application
implementation
The developed application, with the help of
libraries [27,28], consisted of three Google
Colab[29] notebooks and three Python files.
The first notebook is used to translate the
dataset into Ukrainian.
The second notebook is the main neural
network, which is supposed to evaluate the
emotional coloring of messages. First, the
data is obtained from the dataset, the entire
text is divided into training and test data, then
the text is processed (removal of redundant
characters using regular expressions and
lemmatization using the pymorphy2 library).
The neural network is not able to work with
the text, so the text is further tokenized. For
this, all unique words that occur in the dataset
are determined. A dictionary is formed from
them, in which each word is given a unique
token. This allows us to further represent the
text as a list of tokens, where each token
replaces a specific word. The dictionary is
saved in a text file for later use. The data is
also converted to the most suitable data format.
After that, the SentimentRNN neural
network itself is created and trained on
already processed data . Its structure is shown
in Figure 3.
Figure 3. Structure of the SentimentRNN
model
During training, a loss value is calculated
for each epoch, and if it is less than the
previous epochs, then the current state of the
neural network is considered the best. The
best state among all eras is saved to a file for
later use.
The third notebook contains another neural
network. Its task is to determine the intended
scenarios in the user's messages, such as
greetings, farewells, etc. In the data file for
each scenario, certain patterns are defined, as
well as responses to them.
Word processing is similar to the previous
notebook, but word lemmatization is not used
here.
The structure is much simpler as can be
seen in Figure 4.
Figure 4. Structure of the NeuralNet model
Since it is not possible to create
applications with a graphical interface in
Google Colab, this part of the project was
developed in Python IDLE, so we have three
more files with the extension .py.
The first file contains the application
interface, and the second and third are used to
load the two models trained in Google
Colab[29] and saved to files, and also contain
functions for these models to predict the
results.
4. Results of work and testing
If you look at the graphs of training and
validation accuracies and losses, you can see
that the results with the training data were
constantly improving, while the best result on
the validation data was achieved by the model
at epoch 3, after which the losses only
increased. This is indicative of overtraining
and is clearly worth fixing, however 85%
accuracy is a pretty good result, so the
decision was made to leave this trained neural
network as it is for epoch 3.
Then this is used in chatbot. An example of
the conversation with it can be seen in figure 5.
Експертні та інтелектуальні інформаційні системи
27
Figure 5. An example of a conversation with a
developed chat bot
As you can see, the bot can recognize such
common scenarios as greetings, thanks and
goodbyes. When the user asks the bot about
bot itself, it explains that it is an ordinary
artificial intelligence and offers the user to tell
what is on his mind.
The following is an example of what an
average user could write to a bot. In this case,
girl broke up with him, and we see that he is
upset. And the bot, or rather the neural
network behind it, is also able to recognize
negative emotions, so it encourages the
unhappy user in response.
Conclusions
In this article, a recurrent neural network
with modification named long short-term
memory was developed to analyze sentiments
in messages.
A large dataset, 50 thousand comments,
containing different reviews and their
sentiments was collected and annotated to
successfully train and validate the model. It
was also translated into Ukrainian language.
The resulting model demonstrated
accuracy 85% in determining sentiments. One
more small model was developed to improve
make chatbot recognise the intended scenarios
in the user's messages.
And finally, chatbot application designed
like an ordinary chat was created. There user
can communicate with chatbot which uses
pre-trained models to understand how user
feels and responds reespectively. Also it has
some customization such as changing design
and bot behavior.
Further improvement of the model’s
classification quality can be achieved by
collecting a larger and better organised
dataset or by researching other RNN
architectures.
References
1. Rodríguez-Ibánez, M., Casánez-Ventura, A.,
Castejón-Mateos, F., & Cuenca-Jiménez, P.
M. (2023). A review on sentiment analysis
from social media platforms. Expert Systems
with Applications, 119862.
https://doi.org/10.1016/j.eswa.2023.119862
2. Ravi, K., & Ravi, V. (2015). A survey on
opinion mining and sentiment analysis: tasks,
approaches and applications. Knowledge-
based systems. Vol. 89, pp. 14-46.
https://doi.org/10.1016/j.knosys.2015.06.015
3. Birjali, M., Kasri, M., & Beni-Hssane, A.
(2021). A comprehensive survey on sentiment
analysis: Approaches, challenges and trends.
Knowledge-Based Systems. Vol. 226,
107134.
https://doi.org/10.1016/j.knosys.2021.107134
4. Yadav, A., & Vishwakarma, D. K. (2020).
Sentiment analysis using deep learning
architectures: a review. Artificial Intelligence
Review. Vol. 53(6), pp. 4335-4385.
https://doi.org/10.1007/s10462-019-09794-5
5. Do, H. H., Prasad, P. W., Maag, A., &
Alsadoon, A. (2019). Deep learning for
aspect-based sentiment analysis: a
comparative review. Expert systems with
applications. Vol. 118, pp. 272-299.
https://doi.org/10.1016/j.eswa.2018.10.003
6. Sarker, I. H. (2021). Deep learning: a
comprehensive overview on techniques,
taxonomy, applications and research
directions. SN Computer Science. Vol. 2(6),
420. https://doi.org/10.1007/s42979-021-
00815-1
7. LeCun, Y., Bengio, Y., & Hinton, G. (2015).
Deep learning. Nature, 521(7553), pp. 436-
444. https://doi.org/10.1038/nature14539
8. Shymkovych V., Telenyk S., Kravets P.
(2021) Hardware implementation of radial-
Експертні та інтелектуальні інформаційні системи
28
basis neural networks with Gaussian
activation functions on FPGA. Neural
Computing and Applications. vol. 33, no.15,
pp. 9467-9479. https://doi.org/10.1007/
s00521-021-05706-3
9. Harumy, T. F., Zarlis, M., Effendi, S., &
Lidya, M. S. (2021, August). Prediction using
a neural network algorithm approach (a
review). 2021 International Conference on
Software Engineering & Computer Systems
and 4th International Conference on
Computational Science and Information
Management (ICSECS-ICOCSIM), Pekan,
Malaysia, IEEE, pp. 325-330.
https://doi.org/10.1109/ICSECS52883.2021.00
066
10.Shymkovych, Volodymyr, Anatoliy
Doroshenko, Tural Mamedov, and Olena
Yatsenko (2022) Automated Design of an
Artificial Neuron for Field-Programmable Gate
Arrays Based on an Algebra-Algorithmic
Approach. International Scientific Technical
Journal "Problems of Control and Informatics"
vol. 67, no. 5, pp. 61-72.
https://doi.org/10.34229/2786-6505-2022-5-6
11.Perera, N. N., & Ganegoda, G. U. (2023). A
Comprehensive Review on Speech Synthesis
Using Neural-Network Based Approaches.
2023 3rd International Conference on
Advanced Research in Computing (ICARC),
Belihuloya, Sri Lanka, IEEE, pp. 214-219
https://doi.org/10.1109/ICARC57651.2023.10
145741
12.Bezliudnyi Y., Shymkovysh V., Doroshenko
A.( 2021) Convolutional neural network model
and software for classification of typical pests.
Prombles in programming. Vol.4, pp. 95-102.
https://doi.org/10.15407/pp2021.04.095
13.Khurana, D., Koli, A., Khatter, K., & Singh, S.
(2023). Natural language processing: State of
the art, current trends and challenges.
Multimedia tools and applications. Vol. 82(3),
pp. 3713-3744. https://doi.org/10.1007/s11042-
022-13428-4
14.Kravets P., Nevolko P., Shymkovych V.,
Shymkovych L. (2020) Synthesis of High-
Speed Neuro-Fuzzy-Controllers Based on
FPGA. 2020 IEEE 2nd International
Conference on Advanced Trends in
Information Theory (ATIT). pp. 291-295.
https://doi.org/10.1109/ATIT50783.2020.9349
299
15. Chai, J., Zeng, H., Li, A., & Ngai, E. W.
(2021). Deep learning in computer vision: A
critical review of emerging techniques and
application scenarios. Machine Learning with
Applications. Vol. 6, 100134.
https://doi.org/10.1016/j.mlwa.2021.100134
16. Kravets, P., Novatskyi, A., Shymkovych, V.,
Rudakova, A., Lebedenko, Y., Rudakova, H.
Neural Network Model for Laboratory Stand
Control System Controller with Parallel
Mechanisms. Lecture Notes on Data
Engineering and Communications
Technologies. Springer, Cham. 2023. Vol
181. pp. 47-58 https://doi.org/10.1007/978-3-
031-36118-0_5
17. Y.S. Hryhorenko, V.M. Shymkovysh, P.I.
Kravets, A.O. Novatskyi, L.L. Shymkovysh,
A.Yu. Doroshenko. A convolutional neural
network model and software for the
classification of the presence of a medical
mask on the human face. Problems in
programming. 2023. Vol. 2. pp. 59-66.
https://doi.org/10.15407/pp2023.02.059
18. Yu, Y., Si, X., Hu, C., & Zhang, J. (2019). A
review of recurrent neural networks: LSTM
cells and network architectures. Neural
computation. Vol. 31(7), pp. 1235-1270.
https://doi.org/10.1162/neco_a_01199
19. Bezliudnyi Y., Shymkovych V., Kravets P.,
Novatsky A., Shymkovych L. Pro-russian
propaganda recognition and analytics system
based on text classification model and
statistical data processing methods.
Адаптивні системи автоматичного
управління: міжвідомчий науково-
технічний збірник. 2023. № 1 (42), c. 15-31.
https://doi.org/10.20535/1560-
8956.42.2023.278923
20. Hochreiter, S., & Schmidhuber, J. (1997).
Long short-term memory. Neural
computation. Vol. 9(8), pp. 1735-1780.
https://doi.org/10.1162/neco.1997.9.8.1735
21. Van Houdt, G., Mosquera, C., & Nápoles, G.
(2020). A review on the long short-term
memory model. Artificial Intelligence
Review. Vol. 53, pp. 5929-5955
https://doi.org/10.1007/s10462-020-09838-1
22. Kader, N.I.A., Yusof, U.K., Khalid, M.N.A.,
Husain, N.R.N. (2023). A Review of Long
Short-Term Memory Approach for Time
Series Analysis and Forecasting. Lecture
Notes in Networks and Systems. Vol 573.
Springer, Cham. pp. 12-21
https://doi.org/10.1007/978-3-031-20429-6_2
23. Long, F., Zhou, K., & Ou, W. (2019).
Sentiment analysis of text based on
bidirectional LSTM with multi-head attention.
IEEE Access,7, pp. 141960-141969.
https://doi.org/10.1109/ACCESS.2019.29426
14
Експертні та інтелектуальні інформаційні системи
29
24. Olah, C. (2015). Understanding LSTM
networks. URL:
https://colah.github.io/posts/2015-08-
Understanding-LSTMs/
25. R. Yu, S. Liu, X. Wang (2024) Dataset
Distillation: A Comprehensive Review. IEEE
Transactions on Pattern Analysis and Machine
Intelligence. Vol. 46, no. 1, pp. 150-170.
https://doi.org/10.1109/TPAMI.2023.3323376
26. IMDB Dataset of 50K Movie Reviews URL:
https://www.kaggle.com/datasets/lakshmi25n
pathi/imdb-dataset-of-50k-movie-reviews
27. TensorFlow. (n.d.). TensorFlow: An end-to-
end open source machine learning platform.
Retrieved from https://www.tensorflow.org/
28. Keras. (n.d.). Keras: The Python deep learning
API. Retrieved from https://keras.io/
29. Google Colab
https://colab.research.google.com/
Одержано: 16.01.2024
Про авторів:
Кобченко Владислав Русланович,
студент магістр НТУ України
«КПІ імені Ігоря Сікорського»
Шимкович Володимир Миколайович,
кандидат технічних наук, доцент НТУ
України «КПІ імені Ігоря Сікорського».
Кількість наукових публікацій
в українських виданнях – понад 30.
Кількість наукових публікацій
в зарубіжних виданнях – понад 10.
Індекс Хірша – 4. https://orcid.org/0000-
0003-4014-2786
Новацький Анатолій Олександрович,
кандидат технічних наук,
доцент НТУ України
«КПІ імені Ігоря Сікорського».
Кількість наукових публікацій
в українських виданнях – понад 30
Кравець Петро Іванович,
кандидат технічних наук, доцент НТУ
України «КПІ імені Ігоря Сікорського».
Кількість наукових публікацій
в українських виданнях – понад 40.
Кількість наукових публікацій
в зарубіжних виданнях – понад 10.
Індекс Хірша – 4. https://orcid.org/0000-0003-
4632-9832
Шимкович Любов Леонідівна,
асистент кафедри інформаційних систем
та технологій НТУ України
«КПІ імені Ігоря Сікорського».
Кількість наукових публікацій
в українських виданнях – 4.
Кількість наукових публікацій
в зарубіжних виданнях – 1.
https://orcid.org/0000-0002-1291-0373
Дорошенко Анатолій Юхимович,
доктор фізико-математичних наук,
професор, професор НТУ України
«КПІ імені Ігоря Сікорського».
Кількість наукових публікацій
в українських виданнях – понад 200.
Кількість наукових публікацій
в зарубіжних виданнях – понад 90.
Індекс Хірша – 7. http://orcid.org/0000-
0002-8435-1451
Місце роботи авторів:
Національний технічний університет
України «Київський політехнічний
інститут імені Ігоря Сікорського»,
проспект Перемоги 37 та Інститут
програмних систем НАН України, 03187,
м. Київ-187, проспект Академіка
Глушкова, 40.
E-mail:
vladik020402@gmail.com,
v.shymkovych@kpi.ua,
a.novatskyi@kpi.ua,
peter_kravets@yahoo.com,
L.shymkovych@gmail.com,
doroshenkoanatoliy2@gmail.com
|