Article · Bulletin of Shakarim University Technical Sciences · January 25, 2026
DEEP LEARNING-BASED HATE SPEECH DETECTION IN KAZAKH: A HYBRID FRAMEWORK FOR ROBUST TEXT ANALYSIS
Daniyar Sultan, Rustam Abdrakhmanov, Eşref Adalı, Tursinbay Turymbetov, Gulvira Baurzhanovna Bekeshova
Abstract
This study presents a new intelligent system based on deep learning methods for the automatic detection of hate speech in the Kazakh language. Particular attention is paid to the specific nature of Kazakh as a resource-poor language, where the limited linguistic data poses significant challenges in building reliable models. A multilingual data corpus covering a wide range of speech contexts was generated and preprocessed using various online sources-social media, forums, and news portals. To improve efficiency and generalization performance, a hybrid architecture was proposed, including convolutional neural networks (CNNs), bidirectional long short-term memories (BiLSTMs), and Transformer attention mechanisms. An evaluation using precision, recall, F1-criterion, and accuracy metrics demonstrated the superiority of the proposed model over traditional machine learning algorithms. The results of the study make a significant contribution to the development of automatic content moderation systems and promote the creation of a safer, inclusive, and linguistically sensitive digital space for Kazakh-speaking users.
More by Gulvira Bekeshova
- RESULTS OF PILOTING RECOMMENDATIONS FOR DEVELOPING TEACHERS’ DIGITAL COMPETENCE AND AUTOMATED MONITORING BASED ON THE DTDC MODEL2026
- USING GAME THEORY IN WIRELESS NETWORKS FOR NETWORK SECURITYTSARKA Science · 2026
- OPTIMIZING MACHINE LEARNING MODELS FOR FAKE NEWS DETECTION: HYPERPARAMETER SELECTION AND ROC CURVES ANALYSISBulletin of Shakarim University Technical Sciences · 2026
- MULTI-OBJECTIVE EVOLUTIONARY OPTIMIZATION OF SECURITY POLICIES IN SOFTWAREDEFINED NETWORKS (SDN) GIVEN TCAM AND LATENCY CONSTRAINTSBulletin of Shakarim University Technical Sciences · 2026