Wasper
a Bulgarian-language model for detecting propaganda in social media content
DOI:
https://doi.org/10.60054/PEU.2025.12.223-230Keywords:
propaganda detection, digital service act, social media, artificial intelligenceAbstract
This paper introduces WASPer, a classification model designed to detect propaganda in Bulgarian-language social media content. In response to the rising threat of AI-generated disinformation and the regulatory requirements of the EU’s Digital Services Act (DSA), WASPer aims to provide a practical and scalable solution for identifying manipulative narratives online. A thematically diverse dataset was constructed by combining manually annotated organic content and synthetic examples generated with a Bulgarian language model (BgGPT). Each text was human-annotated based on the presence of rhetorical techniques commonly associated with propaganda. The dataset was used to train WASPer (a fine-tuned version of the BgGPT 7B Instruct v0.2 model), achieving an F1 score of 0.853 on the test set. WASPer supports the detection of harmful or misleading content in digital spaces such as comment sections and social media threads, contributing to efforts to meet DSA obligations for transparency and risk mitigation.
References
Alexandrov, A., Raychev, V., Müller, M. N., Zhang, C., Vechev, M., & Toutanova, K. (2024). Mitigating catastrophic forgetting in language transfer via model merging. arXiv preprint arXiv:2407.08699.
Ellul, J. (1973). Propaganda. The formation of men’s attitudes. Vintage Books.
Grootendorst, M. (2022). BERTopic: Neural topic modeling with a class-based TF-IDF procedure. arXiv preprint arXiv:2203.05794.
Hanley, H., & Durumeric, Z. (2023). Machine-made media: Monitoring the mobilization of machine-generated articles on misinformation and mainstream news websites. arXiv 2305.09820.
Jiang, A., Sablayrolles, A., Mensch, A., Bamford, C., Chaplot, D., Casas, D., ... & Sayed, W. (2023). Mistral 7B. arXiv preprint arXiv:2310.06825.
Jowett, G., & O’Donnell, V. (1999). Propaganda and persuasion. Sage.
Jowett, G., & O’Donnell, V. (2018). Propaganda & persuasion (Seventh edition.). SAGE.
Lee, A., & Lee, E. (1937). The fine art of propaganda: A study of Father Coughlin’s speeches. Institute for Propaganda Analysis. Retrieved August 8, 2024, from https://archive.org/details/LeeFineArt.
Pequeño, A. (2024). Russia impersonated Americans using nearly 1,000 fake AI-generated X accounts. Forbes. Retrieved August 20, 2024, from https://www.forbes.com/sites/antoniopequenoiv/2024/07/09/russia-impersonated-americans-using-nearly-1000-fake-ai-generated-x-accounts-feds-allege/.
Piskorski, J. et al. (2023). News categorization, framing and persuasion techniques: Annotation guidelines. European Commission, JRC132862.
Silverman, H. (2011). Reuters: Principles of trust or propaganda? Journal of Applied Business Research; Laramie, 27(6), 93–115.
Torok, R. (2015). Symbiotic radicalisation strategies: Propaganda tools and neuro linguistic programming. In Proceedings of the Australian Security and Intelligence Conference, ASIC ’15 (pp. 58–65).
