| dc.description.abstract |
This study presents a comprehensive analysis of political sentiment within the Amharic-speaking community using advanced deep learning techniques. The dataset comprises 8576 Amharic texts, classified into three sentiment categories: positive (2818), negative (4555), and neutral (1203). The data annotation was conducted by five annotators from the Benishangul-Gumuz Prosperity Party, ensuring the reliability and relevance of the sentiment labels.
Four deep learning models were employed to evaluate their performance in sentiment classification: Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM), and Bidirectional Gated Recurrent Unit (BiGRU). The CNN model achieved the highest accuracy at 0.902, followed by LSTM at 0.896, BiLSTM at 0.895, and BiGRU at 0.883. The superior performance of the CNN model is attributed to its ability to effectively extract local features and patterns within the text, which are crucial for accurate sentiment analysis.
This research advances the field of natural language processing by adapting and applying deep learning models to the Amharic language, which has been underrepresented in sentiment analysis studies. The development of a robust sentiment analysis framework provides a foundation for future studies in political sentiment analysis within the region. Additionally, the findings underscore the importance of context in sentiment analysis, as evidenced by the superior performance of the CNN model.
|
en_US |
