Abstract:
Nowadays the growth of internet technology and smartphones help people to liberally express
their idea about health issues, governmental policies and services, political campaigns, business
and product companies, and public services issues through different social media platforms. Due
to the huge amount of data produced on the internet, identifying them into factual statements and
opinions for sentiment analysis classification is a difficult task. Sentiment analysis, which is a
branch of data mining and natural language processing, examines people's feelings, opinions,
sentiments, evaluations, appraisals, attitudes, and emotions toward entities such as products,
services, organizations, individuals, issues, events, topics, and their attributes. In this study, we
developed a Multi-class Amharic Subjectivity Detection and Sentiment analysis using Machine
Learning Approach with natural language processing techniques. For this study, we have
collected data from Ethiopian Ministry of Health, official Facebook page by using a face pager
tool. We have used a total of 4988 annotated sentences for training and testing purposes. Our
current work focused on first classifying statements as Objective, Subjective, and Non-Related. As
we know for sentiment analysis only subjective statements are required and therefore, here we
take only the subjective statements for the classification. Then after taking the subjective statement,
we classify them into six classes: Hope, Fear, Sadness, Anger, Confusion, and Others. TF-IDF is
used for the feature extraction tasks. Then on extracted feature representation, we have employed
two machine-learning algorithms, Random Forest and Support Vector Machines for classification.
We have undertaken several experiments to determine the best performing model. The Support
vector machine model is best performing with 93.78% accuracy for Subjectivity Detection and the
Random Forest model is best performing with 94.3% accuracy for Multi-Class Sentiment Analysis.