HATE SPEECH AND OFFENSIVE LANGUAGE DETECTION FOR AFAAN OROMO FACEBOOK POSTS AND COMMENTS USING RNN.

Abstract

Social media is changing the face of communication and culture in society around the world. People share their information, feelings, and emotions by using social media platform like Facebook. As social media users increasing day to day, cyber hate and offensive speech’s using social media platform are also increasing rapidly. Social media especially Facebook have a very huge impact on the success or destruction to an individual or groups life. Detecting hate and offensive speech is used for the individuals, group, company as well as the government to make decision and take action on the posts and comments that contains violence contents that cause crime and conflict among the people to be removed. In the proposed work we designed and implemented RNN based hate and offensive speech detection model for Afaan Oromo, based on posts and comments which are available on public Facebook pages in Afaan Oromo. Dataset collected from Facebook public pages was contains domains of politics, religious, and ethnic that contains hate and offensive speech content. We collected 7000 dataset from Facebook public pages and dataset labeled by experts into three classes: hate, offensive and neither. We performed various preprocessing techniques before its feed to the recurrent neural network model. In this study, the recurrent neural network based Long Short Term Memory (LSTM) model developed for the detection of posts and comments contains hate and offensive text on Social media. Additionally word embedding was created by applying the word2vec algorithm with a CBOW model, on a corpus collected from Facebook. The experiment was conducted with LSTM models using 80% of the data set for training and 20% for testing the model and selecting the best combination of hyper-parameters Finally, LSTM-based RNN achieved promising result with accuracy 92% and F-score of 93% to detect posts and comments as hate speech, offensive speech, or neither of them through training. Therefore LSTM model is the best mechanism to detect hate speech and offensive language for Afaan Oromo posts and comment on Facebook.