Using NLP specific tools for non-NLP specific tasks. A web security application

Abstract

In this paper we look at the task of detecting URLs corresponding to infected web pages using Machine Learning and Natural Language Processing specific features. We show that these features render better performance than the previously used hand-crafted lexical features and render similar results to the more expensive host-based features. We also introduce a new adjacent task, that of identifying URLs pointing to the download of portable executable files, and show that our models perform very well on this task too.