Вас приветствует Интернет конференция!
Приветствуйем на нашем сайте
METHODS AND MEANS OF DETECTING FAKE MESSAGES IN SOCIAL NETWORKS
19.10.2024 13:01
[1. Информационные системы и технологии]
Автор: Tokariev Volodymyr, PhD, Associate Professor of the department of Information systems, Simon Kuznets Kharkiv national university of economics; Shao Long Wei, master, department of Information systems, Simon Kuznets Kharkiv national university of economics
ORCID: 0000-0002-7143-6165 Tokariev
In the current digital landscape, the rise of new social media platforms has significantly altered traditional methods of news dissemination and the ways in which people consume news. The lack of effective detection mechanisms for fake news has led to its unchecked spread, posing serious threats to the safety and integrity of the public information space. Therefore, the ability to rapidly detect and curb the dissemination of fake news in these public domains is of critical importance.
While existing detection methods have been somewhat successful, most of these approaches rely on single-modal data, such as text or images, and are therefore unable to effectively capture the semantic features of multiple modalities. Although some multimodal detection methods have emerged, they still fall short in terms of facilitating full interaction and integration between different content modalities. Moreover, these methods often lack the necessary mechanisms to validate and interpret the effectiveness of the extracted features. In light of these limitations, this research focuses on the extraction, fusion, and validation of multimodal features in news content to develop a more robust detection framework. First, it is crucial to investigate the characteristics of fake news, as each modality within the news content presents unique properties. Effective feature extraction is needed for both text and image data. Subsequently, the extracted features from different modalities must be fused in a cross-modal framework to uncover hidden supplementary features. Finally, modifications to the current model architecture are required to validate the fused features, thereby improving the overall interpretability of the model fig.1.
Fig.1.The architecture of Text-CNN.
In summary, the main research objectives of this study are as follows:
- multimodal feature extraction. In fake news detection, the modalities involved include both text and image data, requiring separate feature extraction processes. Given the differences in representation and description between text and images, how can effective features be extracted from these different content modalities? Moreover, given the vast and ever-changing landscape of news events, how can the model extract meaningful features from newly emerging news events when most existing models tend to learn features from specific, previously seen events? To address these issues, this research proposes the use of pre-trained deep learning models with a dual-branch network for extracting both shallow and deep features, resulting in a range of feature vectors at different levels of representation;
- multimodal feature fusion. Current multimodal fake news detection methods often rely on simple concatenation of text and image feature vectors, which fails to effectively capture the complementary and differential information between the modalities. This results in suboptimal detection performance. Therefore, how can the model better fuse multimodal features to fully utilize both complementary and differential information? To solve this issue, the research introduces a combinatory fusion mechanism that employs different fusion strategies to enhance the interaction between features across modalities;
- model structure. Deep learning-based fake news detection models often suffer from a lack of interpretability, particularly when dealing with high-dimensional features. Furthermore, after the textual and visual features are fused, they are directly fed into a detector, which lacks a module to verify the quality of the fused features, thus reducing the interpretability of the model. To address this issue, an improved model structure is proposed which includes a reconstruction module to test the effectiveness of the combined features;
This system will serve as a practical tool to verify the effectiveness of the proposed fake news detection method in real-world applications.
Література:
1. Токарєв В., Ільїна І., Шевченко І., Гриценко І. Про один підхід до рішення асиметричної TSP - задачі при B2C доставках за допомогою платформи "Swarm-bot" - system у фізичному неорганізованому середовищі. Системи управління, навігації та зв'язку, 2023. - №4(74), cc. 110-113.
2. Gryzun L, Tokariev V. Mobile applications design for digital education: IT-students’ engagement experience on conditions of online learning the course "Mobile technologies". Proceedings of the 2nd Workshop on Digital Transformation of Education, co-located with the 18th International Conference on ICT in Education, Research, and Industrial Applications: (ICTERI 2023). CEUR Workshop Proceedings, 2023. Ivano-Frankivsk. Ukraine, рр. 110-123.
3. Koshovyi M. D., Pylypenko O. T., Ilyina I. V., Tokarev V. V. Growing tree method for optimization of multifactorial experiments, Radio Electronics, Computer Science, Control, 2023. - № 3, pp. 55–61.
4. Koshevoy N., Ilina I., Tokariev V., Malkova A., Muratov V. Implementation Of The Gravity Search Method For Optimization By Cost Expenses Of Plans For Multifactorial Experiments. Radioelectronic and Computer Systems, 2023. – Vol. 1(105), pp. 23-32.
5. Ільїна І.В., Токарєв В.В., Яковлєв А.В., Шевченко І.І. Використання системи підтримки прийняття рішень для організації гуманітарної логістики, Системи управління, навігації та зв'язку, 2024, №1(75), cc. 88-91.
Ця робота ліцензується відповідно до Creative Commons Attribution 4.0 International License
Знайшли помилку? Виділіть помилковий текст мишкою і натисніть Ctrl + Enter
Другие научные работы даной секции
Конференции
Конференции 2024
Конференции 2023
Конференции 2022
Конференции 2021
