Development and Evaluation of a Combined Model for Genre Classification of Uzbek Media Texts
Annotatsiya
This paper considers the classification of Uzbek media texts into one of five genres. This problem is relevant in light of a fairly noticeable surge in the development of media and the use of the Internet in Uzbekistan. The study uses a combined Bidirectional Long Short-Term Memory + Conventional neural networks architecture trained on a manually annotated corpus of 2500 news articles divided into five genre categories (Politics, Economy and Business, Culture and Art, Sports and Society). As a result of model training, it was found that the model achieved accuracy, recall and F1 scores of 85%, 84% and 84.5%, respectively, on the validation set. Overall, it should be noted that these results highlight the fairly reliable effectiveness of the chosen neural network architecture for solving the problem. Besides, comparative analysis consists of existed related scientific works, which are conducted within past 1-4 years, which shows relevance of the current work.