Feature-Based Ad-Related Script Detection In JavaScript

As the development of the Internet,online advertisements have become one of the most essential business models.These advertisements on the one hand promote the Internet itself,they release users' private information and disturb their browsing experience on the other hand.In order to shield advertisements shown on the websites,Ad-blocking plug-ins using pattern matching such as Adblock and Ghostery are emerging.These Ad-blockers usually use a constantly updating blacklist to match URLs and element attributes appear in the whole web pages and block them immediately.However,they can do nothing to newly appear ad-related patterns and those being obfuscated.After analyzing the fundamental of the generation of dynamic advertisements,this article proposes that script functions in the third party that execute automatically are generally ad-related.It utilizes Jalangi,a dynamic instrumentation tool in JavaScript,to gain complete function call chains in thses scripts.It firstly binds an attribute that records call chains to JavaScript functions in order to save the whole function call chains.It secondly deals with particular functions such as setTimeout(),set Interval(),call()and apply()to ensure the delivery of chains.It finally uses these chains to identify whether a script loaded in the current webpage is ad-related or not.During the period of classifier construction,this article first analyzes static features in ad-related JavaScript files it found and summarizes static features in JavaScript to distinguish ad-related and non ad-related scripts.Then,it trains the classifier using ad-related features based on ad-related scripts.It finally uses test sets to verify the efficiency of the classifier.Experimental analysis shows 93.86% of accuracy in detecting ad-related scripts.As a result,this classifier can conveniently block ad-related scripts before loading which makes up for Adblockers.