conference paper

The Impact of Stopwords Removal on Disinformation Detection in Ukrainian language during Russian-Ukrainian war

Padalko H.; Chomko V.; Chumachenko D.

Publication year: 2024
Source / journal: ceur workshop proceedings
Country of origin: Canada, Ukraine
Language: English
Authors’ affiliations: National Aerospace University, ?Kharkiv Aviation Institute?, Vadym Manko str., 17, Kharkiv, 61070, Ukraine, University of Waterloo, 200 University Ave W, Waterloo, N2L 3G5, Canada, Centre for International Governance Innovation, 67Erb Str W, Waterloo, N2L 6C2, Canada, Balsillie School of International Affairs, 67Erb Str W, Waterloo, N2L 6C2, Canada
Import source: Scopus export

Abstract

Political disinformation is a growing threat to democracy, particularly in the context of warfare like the full-scale Russian invasion of Ukraine. Analyzing Ukrainian-language disinformation, especially on platforms like Telegram, is essential for understanding the narratives used by hostile actors. This study addresses the gap in Ukrainian-language research by applying advanced topic modelling techniques to improve disinformation analysis. Using a dataset of Ukrainian news articles and titles, we employed the BERTopic model, leveraging BERT-based embeddings and hierarchical clustering. The results showed that topic modelling performs better on full news bodies than titles, and removing stopwords significantly enhances topic clarity. Hierarchical clustering and topic modelling revealed consistent patterns, highlighting the importance of using both methods for comprehensive analysis. This study offers valuable insights into Ukrainian disinformation tactics and methodological improvements for more accurate topic modelling, aiding efforts to counter disinformation in politically sensitive contexts. © 2024 Copyright for this paper by its authors.

Author keywords

bert, disinformation, stopwords, telegram, topic modelling