Штучний інтелект

Анотація: Telegram has emerged as the dominant information platform in Ukraine, where many users rely on it daily as their primary news source, emergency alert system, and government communication channel. Despite this central role in one of the most significant armed conflicts of the twenty-first century, large-scale longitudinal datasets capturing the dynamics of Ukrainian Telegram channels remain largely absent from the research literature, leaving a critical gap in the study of conflict-zone information ecosystems. This paper presents a comprehensive dataset collected from 20 Ukrainian Telegram channels spanning January 2021 to March 2026 — a period encompassing the pre-invasion baseline, the acute shock of Russia's full-scale invasion on February 24, 2022, and the prolonged attritional phase that continues at the time of writing. The channels are distributed across six editorially diverse categories — Official Government, Military, Mainstream Media, Independent Journalist, Anonymous/Aggregator, and Regional — and the dataset captures both posts and comments with full engagement metadata including view counts, forward counts, emoji reactions, and media type classifications. We describe two complementary collection architectures designed for different operational requirements: a local fault-tolerant scraper for initial historical backfill, featuring JSON-based state persistence for crash recovery, batched CSV writing for memory management, and upsert-based deduplication via composite identifiers; and a cloud-native serverless pipeline deployed on Azure Functions for continuous daily synchronization, using StringSession-based stateless authentication and Azure Blob Storage for durable output. Our descriptive analysis across visualizations and tables reveals that the full-scale invasion produced an order-of-magnitude spike in posting volume that permanently elevated the information ecosystem above its pre-war baseline, fundamentally altered media type usage patterns across channel categories, triggered engagement convergence among previously divergent editorial profiles, and created temporal correlation structures among anonymous channels that raise questions about coordinated behavior. We further demonstrate that Telegram covers only 2 of 19 content moderation policy categories identified in prior comprehensive platform analyses — the lowest of any systematically examined platform — creating what we term a "platform moderation vacuum" that makes independent research datasets essential for understanding information dynamics in conflict zones. The dataset, collection pipeline, and analytical notebooks are released to support downstream research in misinformation detection, narrative tracking, sentiment analysis, and computational linguistics for underserved languages.

Ключові слова: social network analysis, social media, content analysis, engagement, data pipeline, Azure, information ecosystem, classification.

Посилання:

1. Arora, A., Nakov, P., Hardalov, M., Sarwar, S. M., Nayak, V., Dinkov, Y., Zlatkova, D., Dent, K., Bhatawdekar, A., Bouchard, G., & Augenstein, I. (2023). Detecting harmful content on online platforms: What platforms need vs. where research efforts go. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (pp. 1–18). Association for Computational Linguistics. https://doi.org/10.48550/arXiv.2103.00153
2. Ostrovska, O., & Lyashkevych, V. (2025). Automated Online Detection of Harmful and Dangerous Content in Social Networks: a Systematic Review. Artificial Intelligence, 30(AI.2025.30(4)), 108–116. https://doi.org/10.15407/jai2025.04.108
3. Paula Fortuna and Sérgio Nunes. 2018. A Survey on Automatic Detection of Hate Speech in Text. ACM Computing Surveys 51, 4, Article 85 (July 2018), 30 pages. https://doi.org/10.1145/3232676
4. Momchil Hardalov, Ashish Arora, Preslav Nakov, and Isabelle Augenstein. 2022. A Survey on Stance Detection for Mis- and Disinformation Identification. In Findings of the Association for Computational Linguistics: NAACL 2022. Association for Computational Linguistics, Seattle, WA, 1259–1277. https://doi.org/10.18653/v1/2022.findings-naacl.94
5. Anna Schmidt and Michael Wiegand. 2017. A Survey on Hate Speech Detection using Natural Language Processing. In Proceedings of the 5th International Workshop on Natural Language Processing for Social Media (SocialNLP '17). Association for Computational Linguistics, Valencia, Spain, 1–10. https://doi.org/10.18653/v1/W17-1101
6. Center for Countering Disinformation of Ukraine. 2023. IPSO Classification and Documented Campaigns Report. Retrieved March 15, 2025 from https://cpd.gov.ua/
7. International Centre for Defence and Security. 2024. Russian Information Warfare Against Ukraine: Analysis of Telegram Operations. ICDS, Tallinn, Estonia. Retrieved March 15, 2025 from https://icds.ee/
8. Ukrinform. 2024. Maidan-3 Operation: Russia Allocates $250M Budget for Telegram Manipulation. Retrieved March 15, 2025 from https://www.ukrinform.net/
9. Jason Baumgartner, Savvas Zannettou, Brian Keegan, Megan Squire, and Jeremy Blackburn. 2020. The Pushshift Telegram Dataset. In Proceedings of the 14th International AAAI Conference on Web and Social Media (ICWSM '20). AAAI Press, Atlanta, GA, 840–847. https://doi.org/10.48550/arXiv.2001.08438
10. Geissler, D., Bär, D., Pröllochs, N., & Feuerriegel, S. (2023). Russian propaganda on social media during the 2022 invasion of Ukraine. EPJ Data Science, 12(1). https://doi.org/10.1140/epjds/s13688-023-00414-5
11. Aleksandra Urman and Stefan Katz. 2022. What They Do in the Shadows: Examining the Far-Right Networks on Telegram. Information, Communication & Society 25, 7 (2022), 904–923. https://doi.org/10.1080/1369118X.2020.1803946
12. Massimo La Morgia, Alessandro Mei, Alberto Maria Sabella, and Francesco Sassi. 2023. Uncovering the Dark Side of Telegram: Fakes, Clones, Scams, and Conspiracy Movements. In Proceedings of the ACM Web Conference 2023 (WWW '23). ACM, New York, NY, 741–752. https://doi.org/10.1145/3543507.3583271
13. Mohamad Hoseini, Philipe Melo, Humberto Benevenuto, Anja Feldmann, and Savvas Zannettou. 2023. Globalization of the QAnon Conspiracy on Telegram. In Proceedings of the 17th International AAAI Conference on Web and Social Media (ICWSM '23). AAAI Press, Limassol, Cyprus, 358–369. https://doi.org/10.1145/3578503.3583603
14. Bertie Vidgen and Leon Derczynski. 2021. Directions in Abusive Language Training Data, a Systematic Review: Garbage In, Garbage Out. PLoS ONE 16, 1 (2021), Article e0243300. https://doi.org/10.1371/journal.pone.0243300
15. Irina Khaldarova and Mervi Pantti. 2016. Fake News: The Narrative Battle over the Ukrainian Conflict. Journalism Practice 10, 7 (2016), 891–901. https://doi.org/10.1080/17512786.2016.1163237
16. Ulises Mejias and Nikolai Vokuev. 2017. Disinformation and the Media: The Case of Russia and Ukraine. Media, Culture & Society 39, 7 (2017), 1027–1042. https://doi.org/10.1177/0163443716686672
17. Maryna Zhdanova and Dariya Orlova. 2019. Computational Propaganda in Ukraine: Caught Between External Threats and Internal Challenges. Working Paper 2019.9. Oxford Internet Institute, Oxford, UK.
18. Samuel Woolley and Philip Howard. 2018. Computational Propaganda: Political Parties, Politicians, and Political Manipulation on Social Media. Oxford University Press, Oxford, UK. https://doi.org/10.1093/oso/9780190931407.001.0001
19. Monastyrskyi, L., & Hura, V. (2024). Utilizing cloud technologies for air quality analysis and monitoring. Electronics and Information Technologies, 24. https://doi.org/10.30970/eli.24.4
20. Hura, V., Monastyrskyi L. (2023). IOT-based solution for detection of air quality using ESP32. Artificial Intelligence, 28(AI.2023.28(3)), 86–93. https://doi.org/10.15407/jai2023.03.086
21. Telethon Contributors. n.d. Telethon Documentation. Retrieved March 15, 2025 from https://docs.telethon.dev/