Detection of offensive content in the Kazakh language using machine learning and deep learning approaches

Area of Science:

• Computational Linguistics
• Artificial Intelligence
• Social Media Analysis

Background:

• Social media platforms face challenges in detecting destructive content like extremism and cyberbullying.
• Standard natural language processing (NLP) models are often inadequate for morphologically rich languages such as Kazakh.
• There is an urgent need for specialized tools to ensure online safety in the Kazakh digital space.

Purpose of the Study:

• To adapt and apply machine learning and deep learning techniques for detecting destructive content in the Kazakh language.
• To address the linguistic complexities of Kazakh, including its agglutinative structure and rich morphology.
• To improve the accuracy and effectiveness of content classification for online safety.

Main Methods:

• Utilized machine learning algorithms including logistic regression and support vector machines (SVM).
• Employed deep learning techniques such as long short-term memory (LSTM) networks.
• Combined n-gram features and stemming methods with machine learning for enhanced classification.

Main Results:

• Achieved high accuracy in classifying destructive content in the Kazakh language.
• Demonstrated the effectiveness of integrating n-gram and stemming with machine learning approaches.
• Validated the performance of adapted NLP models on Kazakh text data.

Conclusions:

• Developing language-specific NLP tools is crucial for handling the complexities of Kazakh.
• The study provides a successful framework for detecting harmful content in lesser-resourced languages.
• Findings contribute to enhancing online safety and combating digital extremism in Kazakh-speaking communities.