Detection of abnormal item based on time intervals for recommender systems.

Area of Science:

• Computer Science
• Information Systems

Background:

• Personalized recommendations are crucial for e-business success.
• Collaborative filtering, a common recommendation method, is vulnerable to shilling attacks.
• Existing shilling attack detection methods face challenges with model dependency and computational cost.

Purpose of the Study:

• To propose an effective and efficient approach for detecting abnormal items caused by shilling attacks.
• To address the limitations of existing methods in terms of attack model dependency and computational expense.

Main Methods:

• Analysis of common features across various shilling attack models.
• Development of a revised bottom-up discretized approach utilizing time intervals and identified features.
• Application of chi-square distribution (χ(2)) to compare rating distributions across different time intervals for anomaly detection.

Main Results:

• The proposed approach demonstrates a high detection rate for shilling attacks.
• The method achieves low computational cost, particularly when the number of attack profiles exceeds 15.
• Efficiency in detecting shilling attacks is improved by narrowing down suspicious users.

Conclusions:

• The developed approach offers a robust solution for identifying abnormal items in recommender systems.
• This method enhances the security and reliability of e-business platforms against shilling attacks.
• The findings contribute to more trustworthy personalized recommendation systems.