Machine Learning-Based Categorization of Racism in Amharic Text Using Multi-Class Classification

The study focuses on combating racism on social media, especially in the Amharic language, a critical issue in Ethiopia due to its widespread online user base. The data collection process utilized tools such as Face Pager and Beautiful Soup. Feature extraction was carried out using word2vec. We develop a classification model to automatically identify and categorize racist content on platforms like Facebook and Telegram. This study aims to categorize the extent of racist content into individual racism, regional racism, country-level racism, and Non-racism. We employed a range of machine learning models, including SVM, Random Forest, and Naive Bayes, alongside deep learning models like RNN, LSTM, and CNN for Multi-Class Classification. These models are rigorously evaluated using standard classification metrics. In comparison to the other models, the Bi-directional Long Short-Term Memory (BI-LSTM) model demonstrated significantly higher accuracy, achieving an impressive 99.68% accuracy rate. This implies that the BI-LSTM model outperformed the remaining models in terms of its ability to accurately classify the data.