COMPARATIVE EVALUATION OF EVIDENCE RETRIEVAL METHODS FOR RAG IN UZBEK- LANGUAGE EDUCATIONAL DOCUMENTS
DOI:
https://doi.org/10.5281/zenodo.23180813Abstract
A question-answering system that uses online educational sources must first retrieve a relevant passage to support its answer. This article examines the retrieval stage of a RAG system working with Uzbek educational documents. The study aims to identify which method places the designated evidence passage among the highest-ranked results in a small collection of Uzbek texts. We extracted 111 passages from eight pages on the gov.uz portal and compared word-level TF-IDF, two character-level TF-IDF variants, and latent semantic analysis (LSA) using 50 questions. TF-IDF based on character sequences of length 3–5 retrieved the designated passage within the first three results for 45 of the 50 questions, compared with 36 for word-level TF-IDF. These findings provide preliminary practical evidence for selecting a retrieval method for Uzbek educational texts. The study evaluates passage retrieval only; it does not assess generated answers or their factual accuracy.Keywords
Uzbek, educational documents, RAG, evidence retrieval, TF-IDF, character n-gramsReferences
1. Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., & Yih, W. (2020). -t., Rocktäschel T., Riedel S., Kiela D. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. Advances in Neural Information Processing Systems, 33.
2. Köse S., K., Baytekin M., C., Aktaş, B., Görür B., K., Munis E., A., Yılmaz, D., Kartal M., Y., & Toraman, C. (2026). RAGTurk: Best Practices for Retrieval Augmented Generation in Turkish. Proceedings of the Second Workshop Natural Language Processing for Turkic Languages (SIGTURK, 179–196. https://doi.org/10.18653/v1/2026.sigturk-1.15
3. scikit-learn. TfidfVectorizer documentation. – https://scikit-learn.org/stable/modules/generated/sklearn.feature_ extraction.text.TfidfVectorizer.html.
4. scikit-learn. TruncatedSVD documentation. – https://scikit-learn.org/stable/modules/generated/sklearn.decomposition. TruncatedSVD.html.
5. Government Portal of the Republic of Uzbekistan. Academic Leave. – https://gov.uz/oz/advice/65/document/1755.
6. Government Portal of the Republic of Uzbekistan. Procedure for Reimbursing Part of Students’ Rental Payments. –
7. Government Portal of the Republic of Uzbekistan. Submitting an Application for Accommodation in Student Dormitories of State Higher Education Institutions. – https://gov.uz/oz/advice/65/document/3106.
8. Government Portal of the Republic of Uzbekistan. Education Loan. – https://gov.uz/oz/advice/65/document/3248.
9. Government Portal of the Republic of Uzbekistan. Assessment of Students’ Knowledge. – https://gov.uz/oz/advice/65/ document/1758.
10. Government Portal of the Republic of Uzbekistan. The Three-Year Work Obligation for Students Who Studied on an Education Grant Has Been Abolished. – https://gov.uz/oz/advice/65/document/3250.
11. Government Portal of the Republic of Uzbekistan. Answers to 7 Important Questions on Transferring Studies. – https:// gov.uz/oz/advice/65/document/4173.
12. Ministry of Digital Technologies of the Republic of Uzbekistan. Students Can Now Apply for Dormitory Accommodation and Temporary Registration via my.gov.uz. – https://gov.uz/oz/digital/news/view/200526.
13. Bobojonova, L., Akhundjanova, A., Ostheimer P., S., & Fellenz, S. (2025). BBPOS: BERT-based Part-of-Speech Tagging for Uzbek. Proceedings of the First Workshop on Language Models for Low-Resource Languages, 287–293.
Downloads
Published
How to Cite
Issue
Section
