Detection Mechanism for Malicious Messages on KSU Student Social Network

Abstract

The internet has a considerable effect on social relations and connections among people. Social networking platforms have been an enormous medium for establishing relations and connections among different people all over the world. People, organizations and companies use these platforms to communicate and interact with their communities and audience. These platforms have made it easy for people to share information, create content, and communicate and connect with others online; however, online interaction and communication among people have resulted in the creation of many problems. Malicious contents can easily be shared and populated to reach a wider audience than by using the traditional sharing methods. Detection mechanism is a growing area of research that can detect any inappropriateness of data that is more sensitive to malicious behavior. The detection mechanism needs to be involved in the analysis of the abusing messages posted on the Twitter account of King Saud University (KSU). Text mining is one approach that can be used to detect such malicious or abusing messages. Text mining techniques provide the means to perform data classification where messages can be classified into malicious and non-malicious messages. In addition, Sentiment Analysis is used to identify user tendencies, trends, and opinions by classifying a text into positive, negative and neutral. In this paper, we aim to provide a literature review to investigate the current techniques. The study also addresses the detection of malicious messages which identifies the behavior of malicious and abusive messages. Based on the extensive review of the current techniques, our focus is on the analysis of Arabic and English tweets on KSU’s Twitter account. First, data was collected from Twitter. This was followed by the preprocessing phase. Then, a corpus was produced applying a machine learning based approach by using Naive Bayes and Random Forest Classifier algorithms. Subsequently, the study focused on comparing the accuracy and performance of the Naive Bayes classifier with Random Forest Classifier algorithms in analyzing Arabic and English texts. In order to ensure reaching accurate results, Arabic and English tweets were analyzed.