A NER-Augmented Deep Learning Framework for Karakalpak Medical Text Classification
Year: 2026
Type: conference-paper
Abstract: Processing medical texts in low-resource Turkic languages is a challenging task due to the limited volume of labeled data, orthographic variability, and morphological complexity. This paper proposes a hybrid approach to classifying medical texts in Karakalpak, combining a script standardization stage, a neural network module for named entity recognition, and a rule-based post-processing algorithm. The BiLSTM-CRF neural network model, designed to extract key medical entities such as diseases, symptoms, medications, and procedures, is used as the structuring layer. At the postprocessing stage, a dictionary-based entity aggregation mechanism is applied, enabling text classification into one of the predefined medical fields based on the accumulation of thematic features. Experimental evaluation confirms the robustness and interpretability of the proposed solution, as well as its suitability for work under resource constraints. The proposed approach can be used in practical medical information systems and adapted to other agglutinative languages with similar limitations. In addition to this, the article contains discussion of relevant works, which are dedicated to text analysis of Turkic languages (including Karakalpak).
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7202264147