You Look Suspicious!!: Leveraging Visible Attributes to Classify Malicious Short URLs on Twitter

Abstract

Twitter is one of the most popular Online Social Networks (OSN). It is used by millions of users worldwide everyday. Due to the text limitation on Twitter (140 characters per tweet), URL (Uniform Resource Locator) shortening services are widely used, however they are not free from risks. Shortened URLs are completely different from the original URLs, and hence users have no idea where the short URLs will direct them to. Attackers leverage this knowledge to their advantage to spread malicious URLs. Most of the approaches proposed for classifying malicious URLs utilize information from both social networks and URL shorting service providers. In this paper, we propose a novel approach to detect malicious short URLs using only visible features of tweets and user profiles. We test four machine learning algorithms, i.e., Naive Bayes, random forest, support vector machine, and logistic regression and obtain an accuracy of up to 97% using random forest when classifying malicious short URLs. Our testing results indicate that the approach using visible features from social networks to detect malicious URLs is practicable.