Fellowship Final Report: Shillings’ Attacks Detection in Recommendation Systems Using Hybrid Adversarial Deep Learning

Recommendation systems are ubiquitous in our daily lives, from suggesting products on e-commerce platforms to recommending movies on streaming services. These systems work by predicting a user’s preferences based on their past behavior and feedback, such as ratings or reviews. However, these systems are vulnerable to malicious attacks, such as shilling attacks, where fake ratings or reviews are injected into the system to manipulate recommendations. Detecting and mitigating such attacks is crucial to ensure the integrity and trustworthiness of recommendation systems. These fake ratings or reviews can be used to promote or demote certain items, leading to biased recommendations. Shilling attacks are challenging to detect because they are designed to mimic the behavior of genuine users, and the fake data is often mixed with genuine data. The problem of shilling attack detection has been studied extensively in the literature, and various methods have been proposed to address this problem. One approach is to use machine learning techniques to identify anomalies in the data, such as unusual rating patterns or review content. Another approach is to use graph-based methods to identify suspicious users or items based on their connections in the network. However, these approaches have their limitations.