TF-IDF Based Classification of Uzbek Educational Texts

This paper presents an approach to automatic Uzbek text classification. Uzbek language is a morphologically rich and low-resource language. The approach integrates Term Frequency–Inverse Document Frequency (TF-IDF) representation with conventional machine learning and similarity-based approaches. The aim is to categorize learning materials at the school grade level to support improved alignment of materials and student learning outcomes. In order to carry out the research, a dataset of 5th-11th grade school textbooks in different subjects was collected. The texts were preprocessed using standard natural language processing (NLP) tools and were transformed into TF-IDF vectors. These were used to train three common classification models: Logistic Regression (LR), k-Nearest Neighbors (k-NN), and Cosine Similarity (CS).Each new input text is compared with the grade-level textbook corpus, and the grade with the highest similarity is selected. It provides an estimate of the appropriate intellectual level for the material. The experimental findings indicate that Logistic Regression achieved 82% accuracy, and Cosine Similarity performed slightly better at 85.7%. Conversely, the k-NN method achieved only 22% accuracy, indicating its low applicability for Uzbek text classification. Overall, the proposed approach demonstrates practical value for pedagogical purposes and potential applicability to wider document analysis issues.