Detection of Hate Speech Text in Afan Oromo Social Media using Machine Learning Approach

Abstract

Objectives: This study aims to develop a hate speech detection model for Afan Oromo’s texts on social networks like Facebook and Twitter using a machine learning algorithm. Methods: we collected comments and posts from social media like Facebook and Twitter pages of BBC Afan Oromo, OBN Afan Oromo, Fana Afan Oromo Program, Politicians, Activists, Religious Men, and Oromia Communication Bureau using Face pager tool. The collected data was labelled using Afan Oromo hate speech evaluation system we developed. Text preprocessing tasks applied on data to remove special characters, stop-words,HTML Tags, extra whitespaces, numbers, lemmatization. The n-gram and TFIDF was applied for feature extraction task to obtain benchmark Afan Oromo hate speech detection dataset. Researchers split dataset into train and test set. Finally, we applied Support Vector Classifier, Multinomial NB, Linear Support Vector Classifier, Logistic Regression decision tree and Random Forest Classifier on 67% of trained data. The performance of proposed model also evaluated using F-score. We also test the performance of developed model by loading test set into it. Findings: Hate speech on social media violates the welfare of Ethnic groups and citizens for living together. Many researches have been doing for English, Amharic, and other Languages to detect hate content from social media. This study has focused on developing a prototype for Afan Oromo hate speech detection model using machine learning algorithms and evaluate its performance in which we found Linear Support Vector Classifier scored highest f1-score value is 64%. Novelty: Afan Oromo hate speech detection framework proposed and successfully implemented to develop Afan Oromo hate speech detection model. We wrote python script that overcome problems typos in Afan Oromo in addition to designing python scripts that recognized apostrophe “ ’ ” as important letter for Afan Oromo word formation. Yet, no researchers have used combination of n-gram and TF-IDF for feature extraction. In this study, the n-gram and TF-IDF used for feature extraction approach to build model that detect Afan Oromo hate speech on Social media. Keywords: Afan Oromo; Decision tree; Facebook; Hate Speech; Linear Support Vector Classifier; Machine Learning; MultinomialNB; Social Media; Support Vector Classifier; Decision Tree and Random Forest Classifier