Abstract
This paper presents a framework for fake news detection based on ontological modeling and semantic annotation of media texts. The study includes a bibliometric analysis of Scopus publications from 2018 to 2026 to identify trends in fake news detection and semantic approaches. The results show a shift from traditional machine learning methods toward transformer-based, graph-based, and semantic models. An adaptive system architecture is proposed, covering data collection, preprocessing, multi-level annotation in Label Studio, knowledge graph construction, model training, inference, and analytics. A formal ontological model was developed to structure the key elements of news texts, including claim, source, evidence, author intent, target audience, and disinformation techniques. The framework supports multilingual processing in Kazakh and Russian. The dataset consists of 5,000 news articles, evenly distributed between fake and real categories and balanced across both languages. Annotation quality was evaluated using Cohen’s Kappa, with values ranging from 0.72 to 0.81, indicating consistent inter-annotator agreement. The proposed approach provides a structured basis for the further development and evaluation of automated fake news detection systems in multilingual environments.