Abstract:
Due to the political situation in Ethiopia people are using social media to interact and express their opinions online. The production of a vast number of user-generated comments is analyzed using sentimental analysis. For the Amharic language, different sentimental analysis approaches have been researched, but currently, with the vast number of comments available, the more complex it is for both lexical and machines to analyze. In the meantime, Deep Learning has provided better performance for classification problems. Deep Learning algorithms can capture complex phenomena. The purpose of this study is to design a sentiment analysis for Amharic language using four deep learning algorithms such as Convolutional Neural Network, Bidirectional Long Short-Term Memory, and Gated Recurrent Unit, and combination of Convolutional Neural Network, which is popular for feature extraction, with Bidirectional Long Short-Term Memory, which provides a memory cell as well as the past and future sequence for the model to learn. Experimental research design is used to structure the study. The collected 5000 comments from Facebook and Twitter are used for the experiment. After the dataset was collected, the dataset was preprocessed and a Word Embedding model was created using Skip-gram, and the output served as the embedding layer for the deep learning model. The dataset was split such that 80% is used for training, 10% for validation, and 10% for testing. The four models were trained using the training data set and evaluated using accuracy, precision, and recall metrics. Experimental result shows that combining Convolutional Neural Network with Bidirectional Long Short-Term Memory registered a promising result with accuracy, precision, and recall of 91.60%, 90.47%, and 93.91% respectively. For this research, two classes for classification were used:(i.e., positive and negative), however, the reality shows that there are neutral comments and comments that are opposing each other. So, further research is needed for hierarchical and multi-class classification
