Застосування технології нейронних мереж для аналізу суспільної думки
The research is devoted to studying and using neural network technologies, in particular algorithms and methods of natural language processing, to increase the efficiency of studying and analyzing public opinion of Ukraine’s partner countries regarding the war in Ukraine. The research involved analy...
Збережено в:
| Дата: | 2024 |
|---|---|
| Автори: | , |
| Формат: | Стаття |
| Мова: | Англійська |
| Опубліковано: |
The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute"
2024
|
| Теми: | |
| Онлайн доступ: | https://journal.iasa.kpi.ua/article/view/296220 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | System research and information technologies |
| Завантажити файл: |  |
Репозитарії
System research and information technologies| _version_ | 1867334441006792704 |
|---|---|
| author | Perevoznyk, Kyrylo Parzhyn, Yurii |
| author_facet | Perevoznyk, Kyrylo Parzhyn, Yurii |
| author_institution_txt_mv | [
{
"author": "Kyrylo Perevoznyk",
"institution": "Educational and Scientific Institute of Computer Science and Information Technology of National Technical University \"Kharkiv Polytechnic Institute\", Kharkiv"
},
{
"author": "Yurii Parzhyn",
"institution": "Educational and Scientific Institute of Computer Science and Information Technology of National Technical University \"Kharkiv Polytechnic Institute\", Kharkiv"
}
] |
| author_sort | Perevoznyk, Kyrylo |
| baseUrl_str | http://journal.iasa.kpi.ua/oai |
| collection | OJS |
| datestamp_date | 2025-02-09T21:55:38Z |
| description | The research is devoted to studying and using neural network technologies, in particular algorithms and methods of natural language processing, to increase the efficiency of studying and analyzing public opinion of Ukraine’s partner countries regarding the war in Ukraine. The research involved analyzing and processing databases consisting of messages about the war in Ukraine on the social network Twitter. The resulting datasets were used to train several neural network models. The best classification results were obtained with the GPT-3.5-turbo model. For a deeper understanding of the results of the public opinion analysis, we created their visualization. The results of the study have shown the high efficiency of the selected solutions. They may be of great practical importance for improving methods of analyzing public opinion and making informed decisions based on a deep understanding of global feedback. |
| doi_str_mv | 10.20535/SRIT.2308-8893.2024.4.08 |
| first_indexed | 2025-07-17T10:28:24Z |
| format | Article |
| fulltext |
Publisher IASA at the Igor Sikorsky Kyiv Polytechnic Institute, 2024
Системні дослідження та інформаційні технології, 2024, № 4 97
UDC 004.8.032.26
DOI: 10.20535/SRIT.2308-8893.2024.4.08
APPLICATION OF NEURAL NETWORK TECHNOLOGY
FOR PUBLIC OPINION ANALYSIS
K.M. PEREVOZNIK, Y.V. PARZHYN
Abstract. The research is devoted to studying and using neural network technolo-
gies, in particular algorithms and methods of natural language processing, to in-
crease the efficiency of studying and analyzing public opinion of Ukraine’s partner
countries regarding the war in Ukraine. The research involved analyzing and proc-
essing databases consisting of messages about the war in Ukraine on the social net-
work Twitter. The resulting datasets were used to train several neural network mod-
els. The best classification results were obtained with the GPT-3.5-turbo model. For
a deeper understanding of the results of the public opinion analysis, we created their
visualization. The results of the study have shown the high efficiency of the selected
solutions. They may be of great practical importance for improving methods of ana-
lyzing public opinion and making informed decisions based on a deep understanding
of global feedback.
Keywords: public opinion, neural networks, natural language processing, large lan-
guage models, social networks, classification.
INTRODUCTION
In the context of war, the Ukrainian government has to make prompt decisions on
how to interact with the governments of partner countries. These decisions should
take into account the public opinion of partner countries in the course of the war
in Ukraine. Therefore, monitoring changes in public opinion and the attitudes of
people in different countries towards this war is a key aspect for further analysis
of the prospects of these countries' assistance to our country, because public opin-
ion can influence both the official positions of partner countries and their leaders.
Of course, public opinion is reflected in social media, which can be used as an
information base for this analysis.
To improve the efficiency of public opinion analysis, artificial neural net-
works can be used, so the purpose of this paper is to study the use of algorithms
and methods of natural language processing based on artificial neural network
models to improve the efficiency of studying and analyzing public opinion in
Ukraine's partner countries.
To achieve this goal, the following tasks were solved: search, analysis, and
preparation of datasets; determination of the dependence of the target variable on
the input data for dataset annotation; dataset markup, cleaning, and verification;
tokenization and vectorization of text; creation of training and test samples; de-
termination of the effectiveness of solving the text classification problem using a
Gaussian naive Bayesian classifier model and a multilayer perceptron to deter-
mine further the effectiveness of using large language models.
K.M. Perevoznik, Y.V. Parzhyn
ISSN 1681–6048 System Research & Information Technologies, 2024, № 4 98
The peculiarity of solving the classification problem in this paper is to iden-
tify the types of classes and data that most affect the classification process. That
is, it is necessary to analyze not the mood of people in its classical psychological
sense based on the color of the message, but to determine which posts and mes-
sages on social media can be classified as useful or not useful for helping
Ukraine. For example, messages about support, assistance, and texts about the
success of our troops will be classified as positive, while posts approving of the
Russian government's statements, which often turn out to be fake news and disin-
formation, and texts supporting the war or Russia will be labeled negative. The
issues and problems of combating disinformation in social networks and news are
discussed in Bergstrom C.T., West J.D. [1].
LITERATURE REVIEW
Literature analysis and selection of technological solutions. Many scientific
works are devoted to the problem of using neural networks for analyzing public
opinion. Thus, the work of Wordliczek Ł. [2] points out the need to use neural
networks for sociological research. The works [3–5] discuss the technical aspects
of using neural networks for analyzing public opinion. Gong M. [3] proposes a
deep neural network (DNN) model for text sentiment classification, which, ac-
cording to the author, makes it possible to better extract local and contextual in-
formation from the text. The article by Yang S. [4] proposes to use of a genetic
algorithm to train a multilayer perceptron to increase the sensitivity of public
opinion forecasting. The article by Chen X. et al [5] proposes the use of hybrid
fuzzy neural networks. It should be said that the use of large language models is
limited by their high computational complexity, which is a common disadvantage
of connectionist neural networks [6].
The software model was developed in Python. All parts of the project,
namely data aggregation, processing, analysis, model development, training, and
testing, were carried out in Jupyter Notebook. The most common Python libraries
were used in the work, such as Tensorflow, Keras, Pandas, Sklearn, RegEx,
NumPy, Matplotlib, Seaborn, WordCloud, as well as the Gensim package for text
processing and topic modeling. In this work, we used the Word2Vec model,
which is one of the most popular models for vectorization and contextual word
association.
We also conducted an experiment using the GPT-3.5-turbo model for text
annotation and classification [7]. The choice of these technological solutions is
due to their popularity, wide functionality, convenient interface, availability of
multifunctional libraries, and support of the developer community. To work with
this model we used the API of the OpenAI platform [8–10]. The use of OpenAI's
GPT models is governed by license agreements, which may differ depending on
whether the intended use is commercial or non-commercial. Users must also
abide by certain terms of use and ethical considerations, such as avoiding harmful
or malicious content and complying with privacy laws. However, these
restrictions do not, in our opinion, prevent this task from being accomplished. It is
likely that using a commercial version of GPT-4 would improve the quality of
content analysis but would incur additional financial costs. There are also open-
source models, such as LLaMA, whose use for this task also requires further
research.
Application of neural network technology for public opinion analysis
Системні дослідження та інформаційні технології, 2024, № 4 99
Data selection. Open datasets were chosen to analyze public opinion on the
war in Ukraine: Unveiling Global Narratives: A Multilingual Twitter Dataset of
News Media on the Russo–Ukrainian Conflict (in the future — Zenodo) [11],
which contains 1.5 million tweets in 60 languages, and the dataset from the Kag-
gle platform — Ukraine Conflict Twitter Dataset (in the future — Tweets) [12],
which contains 44 million tweets until June 2023 related to the war between
Ukraine and Russia.
The Zenodo dataset was selected due to the presence of columns generated
using the RoBERTa model [13]. These columns reflect the essence of the text and
indicate its tendency towards being “in support” or “against” Ukraine, Russia, and
the war in general. Each record in this dataset is represented in JSONL format. It
was decided to use the Tweets dataset for data analysis and classification for the
last two months using the GPT-3 model.
MATERIALS AND METHODS
Project structure and implementation. The project consists of three program
code files. The purpose of each of the development files is shown in detail in Table.
Purpose of development files
File File purpose
PartnerSentiment.ipynb Data preparation and data analysis.
ClassifySentiment.ipynb Processing text data and creating a process
for using the OpenAI API
ModelingSentiment.ipynb Text data processing, text vectorization, and modeling.
Data preparation. The project implementation begins with the creation of
the PartnerSentiment.ipnb file, the main parts of which are provided on the
GitHub platform [20]. All the necessary modules were imported and the function-
ality that iteratively transforms Zenodo data from JSONL to dataframe was im-
plemented. Tweets data was also processed, a data frame was created using Pan-
das, and functionality was developed to concatenate certain columns, filter only
English texts, and extract hashtags from the text using a Regex pattern.
Exploratory Data Analysis. After the data preparation was completed, we
started analyzing it. To do this, we created a new column with a shortened date to
the year-month format and grouped the data by this criterion to see how many
unique text messages there were in each month from the beginning of the war un-
til June 2023. The resulting distribution is shown in Fig. 1.
The next step was to analyze the number of tweets from each country per
month. To do this, we used Regex to reduce all the locations of tweet authors to
the country name, and the rest of the records were removed. The processing re-
sults are shown in Fig. 2.
Next, we analyzed the hashtags that users added to their messages. Using the
WordCloud library, we created word distribution graphs for each month.
The analysis of the information shows that in the first months of the war, the
country received significant support and the war in Ukraine became the most
relevant topic of communication on the world's largest online platforms. It is
worth noting that all war crimes were also noticed and brought to the public.
K.M. Perevoznik, Y.V. Parzhyn
ISSN 1681–6048 System Research & Information Technologies, 2024, № 4 100
A more complete picture of the history of changes in the most popular topics of
discussion can be obtained by creating a hashtag timeline. We also analyzed the
hashtags of users with more than 500 thousand followers. This was done to see
what topics are discussed by popular figures and whether they are in favor of
Ukraine.
EXPERIMENTS
Preparation of datasets. To experiment, we prepared the Zenodo dataset for
training. The entire algorithm was implemented in the ModelingSentiment.ipynb
file, the contents of which are available on GitHub [14]. To determine the target
variables in the datasets that are necessary to solve the classification problem, the
following rules were empirically derived:
Fig. 1. Number of tweets per month
Fig. 2. Number of tweets from each country per month
Application of neural network technology for public opinion analysis
Системні дослідження та інформаційні технології, 2024, № 4 101
IF (A1+A2) > 2,5(B1+B2) → Cl (True);
IF 1,7(A1+A2) < (B1+B2) →C2 (False);
Else → C3 (Unknown).
Where: A1 is the column "This statement is against russia" in the Zenodo
dataset; B1 is the opposite column in terms of content; A2 — is the column "This
statement is in favor of Ukraine" in the Zenodo dataset; B2 — is the opposite col-
umn in terms of content; C1 is a positive class; C2 — negative class; C3 — neu-
tral (undefined) class.
Using the identified class labels, we checked the dataset markup. The
markup results are shown in Fig. 3. Next, we created a function to clean the text
from unwanted characters, and links,
and convert it to lowercase. The text
processing function is shown in Fig. 4.
The text was vectorized using the
Word2Vec model, which was trained
on the full corpus of words after
processing. Before converting the data
to a numerical form, it is worth
looking at the distribution of text
lengths. To do this, we created a
process for counting the number of tokens in a message. The process itself and the
graph showing the distribution of text lengths are shown in Fig. 5.
It was found that almost 900 entries were less than 5 in length and did not
contain much meaning, so they were excluded from the dataset.
Fig. 4. Function for text processing
Fig. 3. Distribution of class labels
Fig. 5. Distribution of text lengths
13
13
K.M. Perevoznik, Y.V. Parzhyn
ISSN 1681–6048 System Research & Information Technologies, 2024, № 4 102
The next step was to create a Word2Vec model by splitting all the values of
the processed text into tokens and feeding them to the model for training. As a
result, the model has a vocabulary of 44351 unique tokens and each vector has
100 values. The training result is shown in Fig. 6.
The resulting dataset was divided into training and validation samples. The
validation data accounted for 20% of the entire corpus. It was also foreseen that
words not contained in the model's dictionary would appear. In such cases, in-
stead of these words, a hundred-dimensional vector filled with values of 0.5 is
added to the sum of vectors. The code for this function is shown in Fig. 7.
Modeling. We first solved the classification problem using a Gaussian naive
Bayesian classifier to evaluate the effectiveness of using large language models.
The classification results are shown in Fig. 8.
In multiclass classification, accuracy is calculated by summing up the num-
ber of correct predictions for each class and dividing it by the total number of
predictions made across all the classes. It is a commonly used metric to evaluate
the performance of models in multiclass problems. In this study, the model's
overall accuracy was calculated using this metric, and it was found to be 45%.
The confusion matrix was also analyzed to understand the model's performance
for each class. It was observed that the model performed well for the positive
class, but it had a high false positive rate for the neutral and negative classes. A
new model with four layers of neurons was developed, which improved the accu-
racy to 57%. However, even with this improvement, the model's accuracy was
still not sufficient for practical applications.
Fig. 6. Word2vec model
Fig. 7. The function of converting texts to vectors
Application of neural network technology for public opinion analysis
Системні дослідження та інформаційні технології, 2024, № 4 103
Use of large language models for annotation. To work with large language
models, we created the ClassifySentiment.ipynb file [14]. For the experiment, we
chose the GPT-3.5-turbo model developed by OpenAI. It can be used using the
API. The following restrictions were also used: data were taken only for May and
June of 2023; duplicates were removed; location values were processed and
tweets from Ukraine were removed.
Next, we created a dictionary called mySentiment and the functionality of
working with the OpenAI API, the implementation of which is also given in [14].
The next step was to create functionality for sending queries to the GPT-3.5-turbo
model, as well as the process of updating the dictionary. As a result, we obtained
a dictionary with 59470 entries annotated with a large language model. They were
saved and compared with the results of the multilayer perceptron. The results of
the comparisons are presented in the file “ModelingSentiment.ipynb” on
GitHub [14].
The GPT-3.5-turbo model demonstrates the best result of all the models
studied. This is an expected result, but it demonstrates that the large language
model GPT-3.5-turbo can effectively solve many problems with a fairly simple
functionality using the OpenAI API.
RESULTS
The final stage of the experiment was the analysis of the results of public opinion
classification based on the GPT-3.5-turbo model. First, the change in the distribu-
tion of forecasts for May and June 2023 was demonstrated. The results are shown
in Fig. 9.
The public opinion score for Ukraine has both positive and negative ratings.
The positive assessments are shown in Fig. 10. The negative, neutral, and positive
grades are labeled as 0, 1, and 2, respectively. Thus, the average score from a
country can range from 0 to 2.
Fig. 8. Confusion matrix of Gaussian naive Bayesian classifier
K.M. Perevoznik, Y.V. Parzhyn
ISSN 1681–6048 System Research & Information Technologies, 2024, № 4 104
To improve the overall sentiment towards Ukraine, we need to analyze the
factors contributing to negative public opinion and eliminate them. We can also
highlight the positive aspects of the country, such as its rich culture, history, pic-
turesque nature, and tourist attractions. By promoting these aspects, we can create
a positive image of the country and improve its sentiment.
DISCUSSION
The results of classification using fairly simple models - Gaussian naive Bayesian
classifier and multilayer perceptron - do not allow us to conclude that they can be
practically used to process large data sets required for public opinion analysis. At
the same time, the results of our work show that large language models with an
Fig. 9. Classification results
Fig. 10. Countries with the most positive attitude toward Ukraine
Application of neural network technology for public opinion analysis
Системні дослідження та інформаційні технології, 2024, № 4 105
open API can be effectively used both for analyzing public opinion and for solv-
ing other applied problems. To obtain qualitative results, an important step was
taken in collecting and preparing a large amount of data, which amounted to an
impressive 44 million tweets from the social network Twitter (X). This amount of
data was one of the key indicators for determining an objective assessment of
public opinion in many partner countries regarding the war in Ukraine.
As part of this work, a visualization of the results of the analysis of public
opinion and sentiment regarding the war in Ukraine was created for a better and
deeper understanding.
CONCLUSIONS
The results of the work show that its goal of studying the effectiveness of using
algorithms and methods of natural language processing based on artificial neural
network models to improve the efficiency of studying and analyzing public opin-
ion in Ukraine's partner countries has been achieved. All the research and devel-
opment tasks were solved, namely: a large amount of data was collected and pre-
pared, which amounted to 44 million messages from the social network Twitter
(X); two models were used to analyze the data: a Gaussian naive Bayesian classi-
fier and a multilayer perceptron, and their comparative analysis was conducted,
which made it possible to find out the advantages and features of each model; an
experiment was conducted using a large language model GPT-3.5-turbo, which
demonstrated the high efficiency of its extraction. In addition, to improve the
evaluation of the analysis results, we created their visualization.
REFERENCES
1. C. Bergstrom, J. West, Calling Bullshit: The Art of Skepticism in a Data-Driven
World. Random House, 2020, 336 p.
2. Ł. Wordliczek, “Neural Networks and Political Science: Testing the Methodological
Frontiers,” Revista de Metodología de Ciencias Sociales, no. 57, pp. 37–62, 2023.
Accessed on: January 3, 2024. [Online]. Available: https://revistas.uned.es/ in-
dex.php/empiria/article/view/36429/27239
3. M. Gong, “Analysis of Internet Public Opinion Popularity Trend Based on a Deep
Neural Network,” Computational Intelligence and Neuroscience, vol. 2022, article
ID 9034773, 2022, 11 p. Accessed on: January 3, 2024. [Online]. Available:
https://www.hindawi.com/journals/cin/2022/9034773/
4. S. Yang, “Analysis of Network Public Opinion in New Media Based on BP Neural
Network Algorithm,” Mobile Information Systems, vol. 2022, article ID 3202099,
2022, 9 p. Accessed on: January 3, 2024. [Online]. Available:
https://www.hindawi.com/journals/misy/2022/3202099/
5. X. Chen, S. Duan, S. Li, et al., “A method of network public opinion prediction
based on the model of grey forecasting and hybrid fuzzy neural network,” Neural
Comput & Applic, vol. 35, pp. 24681–24700, 2023. doi: https://doi.org/10.1007/
s00521-023-08205-9
6. Y. Parzhin, V. Kosenko, A. Podorozhniak, et al., “Detector neural network vs con-
nectionist ANNs,” Neurocomputing, Elsevier, vol. 414, pp. 191–203, 2020. doi:
10.1016/j.neucom.2020.07.025
7. R. Kysliy, “Artificial intelligence in natural language: language models, low-resource
languages and dehumanization in texts,” Forum of GlobalLogic company, 2023.
Accessed on: January 3, 2024. [Online]. Available: https://dou.ua/forums/topic/45583/
K.M. Perevoznik, Y.V. Parzhyn
ISSN 1681–6048 System Research & Information Technologies, 2024, № 4 106
8. OpenAI: Home page of the openAI platform. 2023. Accessed on: January 3, 2024.
[Online]. Available: https://platform.openai.com/overview
9. A. Amri, “OpenAI GPT For Python Developers: The art and science of developing
intelligent apps with OpenAI GPT-3, DALL·E 2, CLIP, and Whisper - Suitable for
learners of all levels,” Leanpub, 2023, 236 p. Accessed on: January 3, 2024.
[Online]. Available: https://leanpub.com/openaigptforpythondevelopers/
10. L. Tunstall, L. Werra, and T. Wolf, Natural Language Processing with Transformers:
Building Language Applications with Hugging Face. O’Reilly Media, Inc., 2022, 406 p.
11. S. Hakimov, G. Cheema, Unveiling Global Narratives: A Multilingual Twitter Data-
set of News Media on the Russo-Ukrainian Conflict. 2023. Accessed on: January 3,
2024. [Online]. Available: https://arxiv.org/pdf/2306.12886.pdf
12. “Ukraine Conflict Twitter Dataset,” Kaggle, 2023. Accessed on: January 3, 2024.
[Online]. Available: https://www.kaggle.com/datasets/bwandowando/ukraine-
russian-crisis-twitter-dataset-1-2-m-rows/data
13. Y. Liu, M. Ott, N. Goyal, et al., “RoBERTa: A Robustly Optimized BERT Pretraining
Approach,” The Hugging Face Сommunity. Accessed on: January 3, 2024. [Online].
Available: https://huggingface.co/docs/transformers/model_doc/roberta
14. K. Perevoznyk, “Social Sentiment Analysis,” GitHub, 2023. Accessed on: January 3,
2024. [Online]. Available: https://github.com/Kyrylopara/Social-Sentiment-Analysis
Received 05.02.2024
INFORMATION ON THE ARTICLE
Kyrylo M. Perevoznyk, ORCID: 0009-0009-2327-1501, Educational and Scientific Insti-
tute of Computer Science and Information Technology of the National Technical Univer-
sity “Kharkiv Polytechnic Institute”, Ukraine, e-mail: kyrylo.perevoznyk@cs.khpi.edu.ua
Yurii V. Parzhyn, ORCID: 0000-0001-5727-1918, Educational and Scientific Institute of
Computer Science and Information Technology of the National Technical University
“Kharkiv Polytechnic Institute”, Ukraine, e-mail: yurii.parzhyn@khpi.edu.ua
ЗАСТОСУВАННЯ ТЕХНОЛОГІЇ НЕЙРОННИХ МЕРЕЖ ДЛЯ АНАЛІЗУ
СУСПІЛЬНОЇ ДУМКИ / К.М. Перевозник, Ю.В. Паржин
Анотація. Присвячено дослідженню та використанню технологій нейронних
мереж, зокрема алгоритмів та методів оброблення природної мови, для підви-
щення ефективності вивчення та аналізу суспільної думки країн-партнерів
України щодо війни в Україні. У ході виконання досліджень проаналізовано й
опрацьовано бази даних, що складалися з повідомлень стосовно війни в Украї-
ні у соціальній мережі Twitter. Отримані датасети використано для навчання
декількох моделей нейронних мереж. Найкращі результати класифікації отри-
мано на моделі GPT-3.5-turbo. Для більш глибокого розуміння результатів ана-
лізу суспільної думки створено їх візуалізацію. Результати дослідження пока-
зали високу ефективність обраних рішень і можуть мати важливу практичну
значущість для поліпшення методів аналізу суспільної думки та прийняття об-
ґрунтованих рішень на основі глибокого розуміння глобальних відгуків.
Ключові слова: суспільна думка, нейронні мережі, оброблення природної мо-
ви, великі мовні моделі, соціальні мережі, класифікація.
Received 05.02.2024
|
| id | journaliasakpiua-article-296220 |
| institution | System research and information technologies |
| keywords_txt_mv | keywords |
| language | English |
| last_indexed | 2025-07-17T10:28:24Z |
| publishDate | 2024 |
| publisher | The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" |
| record_format | ojs |
| resource_txt_mv | journaliasakpiua/0a/a9515b69704dd93ce3bb93a4a10ebb0a.pdf |
| spelling | journaliasakpiua-article-2962202025-02-09T21:55:38Z Application of neural network technology for public opinion analysis Застосування технології нейронних мереж для аналізу суспільної думки Perevoznyk, Kyrylo Parzhyn, Yurii суспільна думка нейронні мережі оброблення природної мови великі мовні моделі соціальні мережі класифікація public opinion neural networks natural language processing large language models social networks classification The research is devoted to studying and using neural network technologies, in particular algorithms and methods of natural language processing, to increase the efficiency of studying and analyzing public opinion of Ukraine’s partner countries regarding the war in Ukraine. The research involved analyzing and processing databases consisting of messages about the war in Ukraine on the social network Twitter. The resulting datasets were used to train several neural network models. The best classification results were obtained with the GPT-3.5-turbo model. For a deeper understanding of the results of the public opinion analysis, we created their visualization. The results of the study have shown the high efficiency of the selected solutions. They may be of great practical importance for improving methods of analyzing public opinion and making informed decisions based on a deep understanding of global feedback. Присвячено дослідженню та використанню технологій нейронних мереж, зокрема алгоритмів та методів оброблення природної мови, для підвищення ефективності вивчення та аналізу суспільної думки країн-партнерів України щодо війни в Україні. У ході виконання досліджень проаналізовано й опрацьовано бази даних, що складалися з повідомлень стосовно війни в Україні у соціальній мережі Twitter. Отримані датасети використано для навчання декількох моделей нейронних мереж. Найкращі результати класифікації отримано на моделі GPT-3.5-turbo. Для більш глибокого розуміння результатів аналізу суспільної думки створено їх візуалізацію. Результати дослідження показали високу ефективність обраних рішень і можуть мати важливу практичну значущість для поліпшення методів аналізу суспільної думки та прийняття обґрунтованих рішень на основі глибокого розуміння глобальних відгуків. The National Technical University of Ukraine "Igor Sikorsky Kyiv Polytechnic Institute" 2024-12-25 Article Article Peer-reviewed Article application/pdf https://journal.iasa.kpi.ua/article/view/296220 10.20535/SRIT.2308-8893.2024.4.08 System research and information technologies; No. 4 (2024); 97-106 Системные исследования и информационные технологии; № 4 (2024); 97-106 Системні дослідження та інформаційні технології; № 4 (2024); 97-106 2308-8893 1681-6048 en https://journal.iasa.kpi.ua/article/view/296220/312907 |
| spellingShingle | суспільна думка нейронні мережі оброблення природної мови великі мовні моделі соціальні мережі класифікація Perevoznyk, Kyrylo Parzhyn, Yurii Застосування технології нейронних мереж для аналізу суспільної думки |
| title | Застосування технології нейронних мереж для аналізу суспільної думки |
| title_alt | Application of neural network technology for public opinion analysis |
| title_full | Застосування технології нейронних мереж для аналізу суспільної думки |
| title_fullStr | Застосування технології нейронних мереж для аналізу суспільної думки |
| title_full_unstemmed | Застосування технології нейронних мереж для аналізу суспільної думки |
| title_short | Застосування технології нейронних мереж для аналізу суспільної думки |
| title_sort | застосування технології нейронних мереж для аналізу суспільної думки |
| topic | суспільна думка нейронні мережі оброблення природної мови великі мовні моделі соціальні мережі класифікація |
| topic_facet | суспільна думка нейронні мережі оброблення природної мови великі мовні моделі соціальні мережі класифікація public opinion neural networks natural language processing large language models social networks classification |
| url | https://journal.iasa.kpi.ua/article/view/296220 |
| work_keys_str_mv | AT perevoznykkyrylo applicationofneuralnetworktechnologyforpublicopinionanalysis AT parzhynyurii applicationofneuralnetworktechnologyforpublicopinionanalysis AT perevoznykkyrylo zastosuvannâtehnologíínejronnihmereždlâanalízususpílʹnoídumki AT parzhynyurii zastosuvannâtehnologíínejronnihmereždlâanalízususpílʹnoídumki |