Educational Text Analysis in Uzbek: Developing an NER Algorithm for Academic and Pedagogical Content

The current research article discusses the development of a named entity recognition algorithm for educational texts in the Uzbek language. The work is based on a custom corpus of 15,000 sentences annotated using the BILOU scheme. The mBERT model with custom settings was used for training, which made it possible to achieve fairly high positive results. The testing results showed a picture where the F1-score of the model was 0.90 for datasets of educational texts, and 0.87 on a mixed corpus. These indicators show that the model copes quite successfully with the tasks of analyzing educational content, despite the fact that the corpus, as well as the model settings, were formed from scratch. In the future, the study can be expanded by including dialect data and using more complex neural network architectures. In addition, it should be emphasized that the authors conducted a fairly detailed comparative analysis of existing similar works, which undoubtedly emphasizes the objectivity of the current work.