Talk · AIP conference proceedings · January 1, 2025

From morphological rules to neural networks: A hybrid framework for medical entity extraction in Karakalpak

Davlatyor B. Mengliev, Nilufar Z. Abdurakhmonova, Hulkar Zokirova, Bahodir B. Ibragimov, Madina Jurakulova, Maftuna Abdunazarova

Nilufar Abdurakhmonova

Open via DOI

Abstract

This paper presents a hybrid method for extracting named entities from medical texts in the Karakalpak language. The approach is based on a rule-oriented method that preprocesses the text in the form of morphological analysis of word forms in the text. This analysis is based on rules and a base of affixes that allow the stemming process to be carried out in order to identify the root of a word or correct misspelled words. After preprocessing, the named entities in the text are directly identified using the multilingual mBERT model. To train this language model, a sample of 5,000 sentences marked using the BIOES scheme was used. The test results showed that the hybrid approach outperforms both rule-based methods without a neural network and neural network solutions without preprocessing. The high score is supported by digital indicators, where the accuracy and recall of the model reached 90% and 90%, respectively, and the F1-measure was about 91%. In addition, the authors conducted a comparative analysis of existing solutions and provided information on the Karakalpak language.

Biomedical Text Mining and OntologiesTopic ModelingAdvanced Text Analysis Techniquesmedical named entity recognitionKarakalpak languagemorphological analysis

From morphological rules to neural networks: A hybrid framework for medical entity extraction in Karakalpak

Abstract

More by Nilufar Abdurakhmonova