Major Advisor: Getachew Mamo (PhD)
Despite the increasing use of social media for information and newsgathering, its nature leads to the emergence and spread of incorrect i.e., information that is unverified at the time of posting, which may cause serious damage to government, markets, and society. The information posted on social media, such as fake news/incorrect information/rumors, individual opinions/feelings, hate speeches, inciting ethnic violence have negative impacts on the political transition and other issues of the country. As a solution to this problem, the study was proposed automating the Afaan Oromo text stance detection model using Supervised ML and feature extraction techniques to build a detection model. To investigated this study two datasets were prepared as train stances and train bodies where train bodies have text body and body Id, as well as train stances, have a headline and body ID. The required data were collected from the Facebook public page only and manually labeled into agree, disagree, discuss, and unrelated to headline depending on text body ID and a dataset that consists of 1692 total Afaan Oromo news text from both datasets were out of this 862 was total stances and 830 was total bodies were prepared. The experiment was conducted by using 1692 of the total datasets. We have applied three supervised machine learning algorithms such as LR, RF, and MNB for classification purposes. The researcher attained an average accuracy of 81%, 69%, and 48%, respectively. Experimental results show that the LR performed better than the other models by outperforming all other classifiers and achieving the best results in terms of improved accuracy (81%). Therefore, to conclude, the Logistic Regression (LR) revealed the best algorithm for stance detection for Afaan Oromo text and classification, which gives a valuable approach for researchers and other users. Keywords: Afaan Oromo, Supervised ML, Stance Detection, social media, Facebook