Shilling attack detection for collaborative recommender systems: a gradient boosting method.

Area of Science:

• Computer Science
• Information Security

Background:

• Collaborative filtering recommendation systems are vulnerable to organized malicious shilling attacks, where fake users manipulate ratings.
• Existing attack detection methods often analyze rating data from a single perspective or use a single classifier, limiting their effectiveness.

Purpose of the Study:

• To propose a novel gradient boosting method, XGB-SAD, for detecting malicious shilling attacks in collaborative filtering systems.
• To address the limitations of existing methods by incorporating a double-view analysis and ensemble learning.

Main Methods:

• Analyzed the rating matrix using a double-view perspective (time and item) to define the TPUS collection.
• Employed eXtreme Gradient Boosting (XGBoost) for iterative optimization and ensemble learning to integrate multiple base classifiers into a strong classifier for attack detection.

Main Results:

• The proposed XGB-SAD method demonstrated superior performance compared to existing methods in detecting small-scale attacks.
• XGB-SAD achieved better overall detection rates, validating its effectiveness in identifying malicious shilling attacks.

Conclusions:

• XGB-SAD offers an effective solution for detecting malicious shilling attacks by leveraging a double-view analysis and gradient boosting.
• The method enhances the stability and security of collaborative filtering recommendation systems against sophisticated manipulation.