Enhancing Multi-Agent Coordination for Real-Time Detection of Obfuscated Shilling Attacks

This research project investigates how agentic AI—autonomous AI agents capable of perceiving, reasoning, coordinating, and acting in dynamic environments—can be advanced to protect recommendation systems against increasingly obfuscated shilling attacks. Shilling attacks occur when malicious actors inject fake profiles, ratings, or reviews to manipulate recommendation outcomes for financial, competitive, or political advantage. As attackers adopt more subtle, disguised, and time-distributed behaviors, traditional detection mechanisms become less effective. Building on the foundational work of Project IT47858, the extension focuses on developing a more robust and collaborative multi-agent defense architecture that not only identifies suspicious user behavior patterns but also infers hidden or obfuscated attack signals through enhanced inter-agent communication and reasoning. The goal is to create an adaptive, real-time agentic AI framework capable of detecting attacks that are intentionally designed to mimic legitimate user activity. Through improved coordination strategies, real-time monitoring pipelines, and extended adversarial simulation environments, the project will enable agents to share insights, cross-validate observations, and dynamically adjust defense decisions to preserve system integrity and user trust. The outcome will be a more resilient, scalable, and intelligent defense layer capable of protecting modern recommendation systems against the next generation of stealthy manipulation threats.

Faculty Supervisor:

Rasha Kashef

Student:

Partner:

Nile University

Discipline:

Computer science

Sector:

Education

University:

Toronto Metropolitan University

Program: