Amharic Political Sentiment Analysis Using Deep Learning Approaches

Abstract This research aims to develop a sentiment analysis system specifically designed for the Amharic language. The study employs four deep learning algorithms to achieve this goal: Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), Gated Recurrent Unit (GRU), and a combination of CNN and BiLSTM. The CNN algorithm is utilized for its effectiveness in extracting relevant features from the input data. By applying filters and pooling operations, the CNN can identify important patterns and structures within the Amharic text. The BiLSTM algorithm is chosen for its ability to process sequential information by considering both past and future contexts. It incorporates a memory cell that enables the model to retain important information and understand the dependencies between different parts of the text. Additionally, the GRU algorithm is employed as it offers similar capabilities to BiLSTM but with fewer computational requirements. This allows for more efficient processing without sacrificing performance. The experimental results obtained from the sentiment analysis system indicate that the combination of CNN and BiLSTM yields promising outcomes. The system achieved an accuracy rate of 91.60%, demonstrating its ability to correctly classify sentiments expressed in Amharic text. Furthermore, the precision rate of 90.47% indicates a high level of accuracy in identifying positive and negative sentiments, while the recall rate of 93.91% suggests that the system effectively captures relevant sentiment instances. In summary, this study successfully designs a sentiment analysis system specifically tailored for the Amharic language. By leveraging the capabilities of deep learning algorithms such as CNN, BiLSTM, and GRU, the system demonstrates strong performance in accurately classifying sentiments expressed in Amharic text, as evidenced by the achieved accuracy, precision, and recall rates.