Multi-Topic Classification of Uzbek Texts Using Rule-Based System and Machine Learning

In this article, the authors consider two approaches to thematic classification of Uzbek texts: the rule-based method and the Naive Bayes algorithm. However, it should be noted that the rule-based approach uses a special dictionary of five thousand words, each of which is assigned weighting coefficients (range from 1 to 3) on different topics. In addition, Naive Bayes was trained using Term Frequency - Inverse Document Frequency vectorization, which takes into account the most frequently occurring words in the corpus. The custom corpus itself consists of two thousand and five hundred sentences, which covers all five thematic categories. The testing showed that the Naive Bayes model achieves 84.5% accuracy on the test sample, while the rule-based system - 78.2%. At the same time, the rule-based algorithm is easier to interpret and edit, while Naive Bayes provides greater flexibility and better generalization ability. Moreover, authors also organized comparative research by describing alternative scientific works of Uzbek researches.