Cyber-Bullying and Cyber-Harassment Detection Using Supervised Machine Learning Techniques in Arabic Social Media Contents
Tarek Kanan, Amal Aldaaja, Bilal Hawashin
Abstract
The social media has provided users with the chance to publish their written and multimedia content and express feelings and emotions about particular subjects via the internet. However, some users have abused these platforms by performing various acts such as Cyber-Bullying and Cyber-Harassment. These phenomena are dangerous and have negative psychological, health, and social effects. Although multiple works have focused on detecting these phenomena on English text, few works studied this phenomenon on Arabic. Moreover, these works used limited number of methods and datasets. Furthermore, there is a lack in Arabic datasets that are concerned with this topic. We propose the use of Machine Learning to detect such negative written acts. We apply various classification algorithms to the dataset, and we use various Arabic Natural Language Processing (NLP) tools. To evaluate the performance of the classifiers, we use Recall, Precision, and F1-Measure. The results show that the Random Forest algorithm yields the highest values of F1-Measure. The same results occurred when no stemming and no stop-word removal are applied. However, when separating datasets into Facebook Posts dataset and Twitter Tweets dataset, SVM gives the highest F1-Measure value. Significant tests were conducted to support our results.