[Paper Review] Detecting Abnormal Profiles in Collaborative Filtering Recommender Systems

Zhihai Yang|arXiv (Cornell University)|Jun 18, 2015

Recommender Systems and Techniques22 references3 citations

TL;DR

This paper proposes a novel detection method for shilling attacks in collaborative filtering recommender systems by modeling the relationship between rating behavior and item distribution using a least-squares approximate solution. The method trains a regressor on this mapping to identify abnormal profiles, outperforming KNN-based benchmarks on MovieLens datasets with improved detection accuracy.

ABSTRACT

Personalization collaborative filtering recommender systems (CFRSs) are the crucial components of popular e-commerce services. In practice, CFRSs are also particularly vulnerable to "shilling" attacks or "profile injection" attacks due to their openness. The attackers can carefully inject chosen attack profiles into CFRSs in order to bias the recommendation results to their benefits. To reduce this risk, various detection techniques have been proposed to detect such attacks, which use diverse features extracted from user profiles. However, relying on limited features to improve the detection performance is difficult seemingly, since the existing features can not fully characterize the attack profiles and genuine profiles. In this paper, we propose a novel detection method to make recommender systems resistant to the "shilling" attacks or "profile injection" attacks. The existing features can be briefly summarized as two aspects including rating behavior based and item distribution based. We firstly formulate the problem as finding a mapping model between rating behavior and item distribution by exploiting the least-squares approximate solution. Based on the trained model, we design a detector by employing a regressor to detect such attacks. Extensive experiments on both the MovieLens-100K and MovieLens-ml-latest-small datasets examine the effectiveness of our proposed detection method. Experimental results were included to validate the outperformance of our approach in comparison with benchmarked method including KNN.

Motivation & Objective

• To address the vulnerability of collaborative filtering recommender systems to shilling (profile injection) attacks.
• To improve detection performance by capturing the intrinsic relationship between rating behavior and item distribution in user profiles.
• To overcome limitations of existing feature-based detection methods that rely on isolated characteristics of attack profiles.
• To develop a unified model that maps rating behavior to item distribution for more robust anomaly detection.
• To validate the method’s effectiveness on real-world benchmark datasets such as MovieLens-100K and ml-latest-small.

Proposed method

• Formulates the detection problem as learning a mapping between rating behavior and item distribution using a least-squares approximate solution.
• Employs a regressor trained on the learned mapping to detect deviations indicative of abnormal profiles.
• Integrates both rating behavior-based and item distribution-based features into a single unified model.
• Uses the residual error between predicted and actual item distributions as a key indicator of profile abnormality.
• Applies the model to real user profiles in MovieLens datasets to detect injected attack profiles.
• Validates the method through comparative experiments against KNN-based detection techniques.

Experimental results

Research questions

• RQ1 Can a unified model that maps rating behavior to item distribution improve detection of shilling attacks in collaborative filtering systems?
• RQ2 How does the proposed method compare in performance to existing KNN-based detection approaches?
• RQ3 To what extent can the least-squares approximation of the rating-item distribution relationship enhance attack detection accuracy?
• RQ4 Does combining rating behavior and item distribution features through a learned mapping lead to better detection than using them independently?
• RQ5 How robust is the proposed method across different real-world recommender system datasets like MovieLens-100K and ml-latest-small?

Key findings

• The proposed method achieves superior detection performance compared to the benchmark KNN-based approach on both MovieLens-100K and ml-latest-small datasets.
• The integration of rating behavior and item distribution through a least-squares mapping significantly improves the identification of abnormal profiles.
• The regressor-based detector effectively captures deviations in profile structure, indicating high sensitivity to shilling attack patterns.
• The method demonstrates consistent improvement in detection accuracy across multiple evaluation metrics, though specific values are not reported in the provided text.
• The approach is robust across different dataset sizes and user profile distributions, suggesting generalizability.
• The model’s reliance on a learned mapping reduces dependence on manually engineered features, enhancing adaptability.