Abstract
This study addresses the urgent need to analyze digital threats in everyday discourse by constructing a 1,000text annotated corpus from social media and news platforms covering military and geopolitical events. The purpose of the proposed study is to address the urgent need to analyze digital threats in everyday discourse by creating an annotated corpus of texts with elements of information operations. A multi-layered annotation scheme captures semantic actors and pragmatic features – including impact type, emotional tone, disinformation markers, and intent (e.g., provocation, intimidation). Annotation via Label Studio ensured flexibility, quality control, and context sensitivity, with inter-annotator reliability (Cohen’s Kappa = 0.82) confirming consistency. In pilot experiments, the Onto-IO-BERT model achieved an F1-score of 0.81, outperforming baseline classifiers. Practical utility was validated through analysis of real Telegram messages. The framework is tailored for studying military information operations within Kazakhstan’s Ministry of Internal Affairs and the created corps is a new resource for analyzing military information operations, filling a significant gap in the existing data set. The presented corpus contains texts in Kazakh, Russian and English. The corpus is openly accessible at: https://github.com/baiangali/multi_mil