Identifying Sources and Participants of Propaganda in TikTok Using Machine Learning

The paper presents an approach to identifying sources of disinformation, fakes, and propaganda on the TikTok social network using modern natural language processing (NLP) and artificial intelligence methods. The main goal of the research is to create a system capable of automatically analyzing comments on videos with propaganda content, as well as the source of their distribution and potential participants in propaganda. As part of the research, a corpus of comments in Ukrainian and Russian was manually collected, which were classified as propaganda or neutral. Based on the analysis of the dataset, certain criteria were identified for identifying sources of disinformation and its potential participants, in particular through the use of the Russian language, repeated propaganda narratives, as well as the fakeness of accounts. The steps of the comment preprocessing algorithm are given. Two approaches to analysis were developed: a classification model based on RandomForestClassifier and a clustering model using the KMeans algorithm. Both models use RoBERTa transformers for the respective languages, as well as an additional manually generated set of comment features. A Telegram bot and a graphical interface were built for the convenient use of the system, which allows receiving comments from TikTok, classifying them, and providing the user with analysis results. The proposed system is a relevant tool for information security and combating propaganda in the digital environment.