A new approach for the detection and analysis of phishing in social networks: the case of Twitter

Abstract

Cybercriminals use Internet and social networks as a vector to launch phishing attacks in order to lure victims to disclose personal information. There are several methods for the detection and analysis of these attacks, among the most used are those based on machine learning. However, these methods suffer from a lack of precision in detecting phishing attacks. Therefore, it is necessary to propose new methods to improve the predictions of these attacks. In this article, we propone a three (03) step approach for the detection and analysis of phishing on Twitter, which can be applied to several other social networks. The first step is to browse a database called “Blacklist” in search of the suspicious URL (Uniform Resource Locator). Subsequently, we proceed to URL (Uniform Resource Locator) analysis leveraging machine learning techniques introducing new features. In this step the three following classifiers were used namely, Regression Logistics, SVM (Support vector Machine) and Random Forest. Afterwards, we added a module to analyze Twitter accounts, also based on machine learning, using user-related features to detect malicious twitters who are causing the phishing attacks. Finally, we tested our system on real data and then implemented it in the form of an application for the end users.