Detecting Potential Phishing Attacks on Twitter Using Transformers

Phishing is a social engineering attack where criminals use URLs to deceive users to obtain their login credentials, and personal data, and get unauthorized access to a user’s accounts or gadgets. Therefore, checking the legitimacy of URLs can play a big role in detecting phishing attacks. There are many methods that can achieve accurate phishing detection; the most popular method being machine learning and black-listing or white-listing approaches; leaving other methods like semantic analysis or deep learning approaches unexplored. This paper aims to propose a novel method capable of doing real-time detection and an open-source library that is a curated collection of models that is already made by and is available for everyone in the community. The novel transformers method consists of pre-trained transformers and a black-listing approach which will be combined to increase the efficiency and performance of detection before it can be used properly to classify phishing URLs on Twitter in real-time. This approach reached 97.9% accuracy using a balanced dataset, suggesting that transformers can be a powerful tool in counteracting phishing attacks.