Assessing the Stability and Robustness of Semantic Web Services Recommendation Algorithms Under Profile Injection Attacks

Abstract

Recommendation systems based on collaborative filtering are open by nature, what makes them vulnerable to profile injection attacks that insert biased evaluations in the system database in order to manipulate recommendations. In this paper we evaluate the stability and robustness of collaborative filtering algorithms applied to semantic web services recommendation when submitted to random and segment profile injection attacks. We evaluated four algorithms: (1) IMEAN, that makes predictions using the average of the evaluations received by the target item; (2) UMEAN, that makes predictions using the average of the evaluation made by the target user; (3) an algorithm based on the k-nearest neighbor (k-NN) method and (4), an algorithm based on the k-means clustering method.The experiments showed that the UMEAN algorithm is not affected by the attacks and that IMEAN is the most vulnerable of all algorithms tested. Nevertheless, both UMEAN and IMEAN have little practical application due to the low precision of their predictions. Among the algorithms with intermediate tolerance to attacks but with good prediction performance, the algorithm based on k-nn proved to be more robust and stable than the algorithm based on k-means.

Keywords

• Profile injection attack
• collaborative filtering algorithms
• semantic web services