Talk · International Conference on Software and Computer Applications · February 20, 2025

USING KAZAKH NER DATASETS FOR MULTICLASS CLASSIFICATION IN THE LEGAL DOMAIN: A COMPARATIVE STUDY OF BERT, GPT, AND LSTM MODELS

Дина Оралбекова, Orken J. Mamyrbayev, Ainur Аkhmediyarova, Dinara Kassymova, Zhibek Alibiyeva

Abstract

This study presents an in-depth comparative analysis of the performance of three key approaches in natural language processing (NLP) — transformers, recurrent neural networks (RNNs), and traditional machine learning methods — in the task of multiclass text classification within the legal domain in the Kazakh language. A specialized dataset for named entity recognition (NER), adapted to legal topics, was utilized for the analysis. The primary focus is on the classification of legal texts. Standard evaluation metrics, such as accuracy, recall, precision, and the area under the curve (AUC), were employed to provide an objective assessment of the models' efficiency in text classification tasks. Special attention was given to the processing of the Kazakh language, which belongs to the category of under-researched languages in computational linguistics. This highlights the necessity of developing specialized algorithms and adapting existing methods to work effectively with this language and its legal terminology. The conducted research not only expands the understanding of the capabilities of existing models for text processing in low-resource languages but also emphasizes the importance of further efforts to automate legal services. This can facilitate the development of more accessible and scalable tools for legal assistance.

Topic ModelingArtificial Intelligence in LawText and Document Classification TechnologiesKazakh languagelegal text classificationmulti-class text classification

USING KAZAKH NER DATASETS FOR MULTICLASS CLASSIFICATION IN THE LEGAL DOMAIN: A COMPARATIVE STUDY OF BERT, GPT, AND LSTM MODELS

Abstract

More by Orken Mamyrbayev