Sentiment Analysis on Amharic Language-Based COVID-19 Discourse from Facebook social media comments

Abstract

The new coronavirus disease (COVID-19) outbreak from Wuhan, China in late December 2019. The virus causes respiratory infections ranging from the common cold to more serious respiratory problems. covid-19 pandemic made huge impacts on different sectors environmental, mental, economical, and industrial are some of them which the pandemic affects negatively. prior studies indicated that social media is a key tool used for gaining a huge amount of people’s opinions or sentiments towards such pandemics. Sentiment analysis is an important tool when it comes to analyzing people's expressions and thoughts on social media. The collected sentiments can be very crucial to assist public health authorities in monitoring and tracking of health information, worries, behaviors, and misinformation, and designing interventions to reduce the impact of the pandemic. In such cases, there is a need to develop a system that detects people’s opinion automatically and categorizes them as positive or negative to the guidelines given by health authorities. However, despite the importance of sentiment analysis, much investigation is not done to assess and find people’s attitudes on social media in the context of local Amharic language. The objective of this thesis is to apply sentiment analysis on Facebook social media by extracting Amharic textual comments focuses on Covid-19 and compare the performance of machine learning algorithms to find the best model. In this study, 15,000 comments regarding Covid-19 was collected and 7309 comments extracted during pre-processing stage, after which three supervised machine learning algorithms SVM, Nave Bayes, and Maximum Entropy used with feature extraction BOW, TF-IDF, and word2vec to classify sentiments expressed on comments. From which, Naïve Bayes with TF-IDF yields high results in classifying sentiments with 83.3% accuracy. The experimental evaluation shows how the proposed approach is effective.