AFAN OROMO POLITICAL STANCE CLASSIFICATION USING DEEP LEARNING APPROCHES

Stance classification in Afan Oromo, an important language spoken in Ethiopia, is a challenging task with significant implications for understanding political sentiments. In this study, we propose a deep learning-based approach for Afan Oromo stance classification, leveraging Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Bidirectional LSTM (Bi-LSTM) models. We curated a comprehensive dataset of Afan Oromo text documents, annotated with favor and against stance labels. The dataset was preprocessed to convert the text into numerical representations suitable for deep learning models. We conducted single-task learning experiments, training each model separately for stance classification.our results indicate that the CNN model achieved the highest accuracy of 85.9%, surpassing both LSTM (82.9%) and Bi-LSTM (78.8%) in single-task learning. This highlights the CNN model's ability to capture local patterns and extract relevant features from Afan Oromo text, establishing it as the best-performing model in this study. Additionally, we explored multi-task learning, training the Bi-LSTM and LSTM models on related tasks. In the multi-task learning setting, the Bi-LSTM model achieved an accuracy of 80.9%, while the LSTM model achieved an accuracy of 80.4%. These results demonstrate the challenges and complexities associated with multi-task learning for Afan Oromo stance classification. the findings contribute to sentiment analysis and political discourse analysis in the Afan Oromo language. Accurate stance classification using deep learning models enables applications such as opinion mining, political analysis, and decisionmaking support.