Skip to content
DSML.KZ
Publications

Article · Bulletin of D Serikbayev EKTU · June 30, 2025

A COMPARATIVE STUDY OF LSTM AND BERT MODELS FOR MULTI-CLASSIFICATION TASKS USING NER DATASET

Дина Оралбекова, Orken J. Mamyrbayev, Ayagoz Imansakipova, Aliya Zhiger, Kuralay Mukhsina

Orken Mamyrbayev
Open via DOIPDF

Abstract

The article presents a comparative analysis of LSTM and BERT models applied to multi- classification tasks in the Kazakh language using a named entity recognition dataset. The study primarily focuses on addressing the issue of limited resources for processing Kazakh text by adapting existing machine learning methods for the analysis of multidimensional classification tasks. Both approaches have demonstrated their effectiveness in various aspects of text data processing, including modeling contextual dependencies and accurately classifying multiple categories. The LSTM model exhibited a high capability for capturing temporal dependencies in text, making it suitable for classification tasks in low-resource language settings. Meanwhile, the BERT model, based on the Transformer architecture, showed superior results in contextual analysis and processing of complex text structures, ensuring higher performance in multi-classification of Kazakh text. The experimental results indicate that both models can be effectively employed for text classification tasks in the Kazakh language. However, the BERT model demonstrated more stable and reliable outcomes, attributed to its ability for deeper contextual understanding. The findings underscore the importance of applying modern natural language processing (NLP) methods to low-resource languages and open new avenues for further research and practical application

Topic ModelingNatural Language Processing TechniquesSentiment Analysis and Opinion Miningnamed entity recognitionKazakh languagemulti-class text classification

More by Orken Mamyrbayev