Semantic Representation Learning using Transformer Architectures for Context Aware Detection of Web Application Attack

Abstract

The modern day digital world is virtually entirely run on web applications, which can be used in social media and banking, e-commerce and even healthcare. However, they also continue to get increasingly targetable to advanced cyberattacks such as SQL injection and cross site scripting and command based injection as they get more and more complicated. Old school protection mechanisms such as rule based web application firewalls (WAFs) and other old school security mechanisms are usually ineffective due to reliance with changing, obfuscated or zero day attack. This is why the current work offers a semantic and context aware detection system, which is based on transformer architecture i.e., on Bidirectional Encoder Representations of Transformer architecture. This entire workflow consist of pre processed data, textual feature extraction, transformer based model training and benchmark set data collections. This is because the model is better able to generalize by using real world traffic conditions an not predetermined set of rules, false positive minimized and pattern known and unknown attacks are identified. This method of semantic learning, even when compared with conventional methods, actually increases the detection accuracy, adaptability and scalability