NEURAL NETWORK MODEL FOR AUTOMATIC DETECTION OF KAZAKH-LANGUAGE HATE SPEECH
This article describes how to use modern machine and deep learning to automatically detect hate speech in Kazakh-language web material. The project aims to secure the digital environment and reduce inappropriate language on social media and user platforms. To achieve this goal, methods were developed and tested for data gathering, preprocessing, classification model creation, and optimization. The works collected data from social media (Instagram, TikTok, YouTube, Facebook), news websites, and forums with Kazakh-language user-generated content. Data preprocessing included noise removal, tokenization, lemmatization, and stop word removal. To ensure result reliability, an annotated corpus classified communications as "hate speech" or "non-hate speech". The authors uses hybrid deep learning architectures to construct an efficient and flexible system for automatically identifying harmful language on the Kazakh-language Internet. CNN, LSTM, Bi-LSTM, and Transformer models and their attention process integrations are examined to consider local and contextual text features. The article aims to train and verify a model that can recognize hateful and filthy words in real time. This approach to social network moderation platforms is projected to improve user safety, eliminate dangerous content, and lay the groundwork for Kazakh language natural language processing research. Multilingual platforms can use the proposed method to construct intelligent systems for automatic moderation and long-term digital environment protection.
Сведения
- Опубликовано
- 2025 · № 4
- Язык
- Английский
- ISSN
- 1991-346X · 2518-1726
- Издатель
- Central Asian Academic Research Center LLP