SENTIMENT ANALYSIS OF AFAAN OROMO FACEBOOK COMMENTS ON GOVERNMENTS CURRENT AGENDAS USING DEEP LEARNING APPROACH

Major Advisor: Wakgari Dibaba (Assistant Professor) Abstract This thesis gives a thorough investigation into sentiment analysis utilizing a deep learning approach, focusing on Afaan Oromo Facebook comments pertaining to the current agendas of government officials. Social media has become an essential platform for expressing public opinions and sentiments on various matters, including political issues and government policies. The primary objective of this research is to develop efficient deep learning model capable of accurately discerning the sentiment behind Afaan Oromo Facebook comments related to the ongoing priorities of government officials posts. The important and crucial work in this thesis is to create dataset or corpora from the scratch, because there is no sufficient dataset appropriate for the research. The proposed approach involves pre-processing the textual data to handle language-specific challenges and convert it into a format suitable for deep learning algorithms. The research aims to contribute to the fields of sentiment analysis and NLP by exploring the nuances of Afaan Oromo language in the context of social media. By identifying sentiments expressed by users towards the government officials' current agendas which [are] posted on their official Facebook pages, the comments are taken from Facebook via application called “Apify “. By copying the user’s link to their profile and pasting on Apify application we downloaded the comments along with post title. We have used Dr.Abiy Ahmed, Adanech Abebie, Shimelis Abdisa, Prosperity party, Communication bureau,Taye Dandea official Facebook page. Generally, in this scenario, the Feed forward Neural Network (FNN) emerges as the preferred deep learning model for sentiment analysis based on the experimental findings and the particular characteristics of Afaan Oromoo Facebook data. When compared to LSTM, BiLSTM, and CNN, the FNN model performed better in terms of accuracy, precision, recall, andF1-score. Then by using 13500 data which is preprocessed and cleaned, we got Test Accuracy of 82.3%, 81.6%, 80.9%, and 84.7% for LSTM, BiLSTM, CNN and FNN respectively and depending on our result FNN is the best model for sentiment analysis for Afaan Oromo Facebook data. Depending on our experiment the sentiment of the community towards the government is positive. Key words: - sentiment Analysis, Afaan Oromoo, Deep learning