EXPERIMENTAL ANALYSIS OF MACHINE LEARNING AND DEEP LEARNING MODELS FOR SENTIMENT ANALYSIS IN THE UZBEK LANGUAGE

Abstract

This paper investigates the problem of sentiment detection in Uzbek social media texts through a comparative evaluation of classical and deep learning models. Within the scope of the study, the performance of Naive Bayes, Support Vector Machine (SVM), LSTM recurrent neural network, and transformer-based models such as BERT/RoBERTa was experimentally assessed. The models were trained on a balanced corpus and evaluated using accuracy and F1-score metrics. The results demonstrate that transformer-based models, which effectively capture contextual dependencies, achieved the highest performance, while the LSTM model produced slightly lower but still competitive results. Among traditional approaches, SVM proved to be a stable and effective baseline classifier. The findings confirm the superiority of deep contextual models for Uzbek sentiment analysis, while also indicating that classical or hybrid approaches remain practically relevant for real-time applications.