A novel Kalman Filter based shilling attack detection algorithm

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• Collaborative filtering is a common recommendation system technique.
• Recommendation systems are vulnerable to shilling attacks where fake profiles manipulate item recommendations.
• Existing methods struggle to effectively detect these sophisticated attacks.

Purpose of the Study:

• To propose a novel Kalman filter-based model for detecting shilling attacks in collaborative filtering recommendation systems.
• To identify abnormal rating patterns and suspicious users indicative of malicious activity.
• To enhance the robustness and trustworthiness of recommendation systems against manipulation.

Main Methods:

• A Kalman filter is employed to predict user ratings.
• The model statistically analyzes the difference between actual and predicted ratings to detect anomalies.
• Suspicious ratings are filtered based on identified abnormal time periods, and suspicious users are flagged.

Main Results:

• The proposed Kalman filter-based method demonstrates superior performance in detecting shilling attacks compared to traditional approaches.
• The model effectively identifies abnormal time periods and flags suspicious users.
• Experimental results validate the enhanced detection capabilities of the new method.

Conclusions:

• The Kalman filter-based attack detection model offers a significant improvement for securing recommendation systems.
• This method provides a robust solution for identifying and mitigating shilling attacks.
• The findings contribute to building more reliable and trustworthy recommender platforms.