KazFakeCorpus - A Bilingual Kazakh-Russian Corpus for Fake News Detection with Multi-Level Annotation
Year: 2026
Type: conference-paper
Abstract: The rapid spread of disinformation in digital media makes automatic fake news detection an important task of natural language processing, especially for low-resource and bilingual language environments. This paper presents KazFakeCorpus, a bilingual corpus of news texts in Kazakh and Russian designed for research on automatic fake news detection. The corpus contains 4276 texts annotated using a multi-level scheme that captures not only the veracity of the message but also the type of fake content, disinformation techniques, modality, and source characteristics. REAL texts were collected from the official government portal Gov.kz, while part of the FAKE class was created through controlled synthetic generation based on real news articles followed by expert verification. In the experimental study, classical machine learning methods, neural architectures, and transformer models were compared, including Logistic Regression, Random Forest, CNN, Bi-LSTM, BERT, and XLM-R. The results show that transformer models achieve the best performance, with XLM-R obtaining the highest F1-score, demonstrating the effectiveness of multilingual contextual representations for analyzing bilingual news texts. At the same time, the results should be interpreted considering the limitations of the corpus related to the use of synthetic data, possible generation artifacts, and stylistic bias toward official discourse. The proposed corpus represents a foundational resource for further research in bilingual misinformation detection and the development of explainable, multimodal, and LLM-based approaches.
Institution: L. N. Gumilyov Eurasian National University
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7167253416