Detecting Arabic terrorism messages on twitter using machine learning / Alharbi Norah Muteb S

Alharbi, Norah Muteb S (2019) Detecting Arabic terrorism messages on twitter using machine learning / Alharbi Norah Muteb S. Masters thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (194Kb)
	PDF (Thesis M.A.) Download (1043Kb)

Abstract

Terrorist groups like ISIS are spreading online propaganda using numerous social media platforms such as Twitter and Facebook. Radical groups in the Arab world are making use of these platforms at alarming rates. One of the more common approaches for stopping the use of social media by these terrorist groups involves suspending their accounts once they are discovered. However, the use of this approach requires that analysts manually read and analyze social media activities, which often involve the manual analysis of significant amounts of information. In addition, the existing works are not efficient enough to stop these malicious activities due to lack of research on the retrieval and data mining of data in Arabic, especially those involved in terrorist activities. This research is undertaken to propose an effective text classifier based on machine learning model for detecting terrorism in Twitter, it is an attempt at using a machine learning model that will automatically detect Arabic tweets from terrorist groups on the Twitter platform. Machine learning was used to aid in both the detection and categorization of a set of diverse Arabic tweets. These tweets were finally classified as either radical or not radical. This work has investigated the use of use of five text classifiers , due to the variety of philosophies behind each method and its learning process, to select the best classifier for our features. These classifiers are the Support Vector Machine (SVM), AdaBoost (discrete), AdaBoost (real), Logistic Regression and Naïve Bayes. The performance of these models was evaluated using precision, recall, F-measure, and accuracy. The work produced promising results that suggest that the use of machine learning models to detect radical Arabic content on social media platforms may be used with great potential for yielding results. The experimental results and data suggest that the use of these models yield high accuracy, with the linear classifier yielding the best results with 99.7% accuracy.

Actions (For repository staff only : Login required)