Disclosed herein are enhancements for operating a communication network to detect malware in scripts of web applications. In one implementation, a method for modeling the structure of embedded unclassified scripts to compare the abstract dynamism of similar scripts. The method may determine structure of unclassified end user browser script by building abstract structure using code from unclassified end user browser script; compare determined structure of unclassified end user browser script with a plurality of generalized abstract structures; if the determined structure of unclassified end user browser script matches within a predetermined threshold of any of the plurality of generalized abstract structures, then the unclassified end user browser script is classified as benign, otherwise the determined structure is classified as malicious. This, in turn, provides a scalable and efficient way of identifying benign, malicious, known and unknown scripts from a script available in full or in part.