The Performance of Location Aware Shilling Attacks in Web Service Recommendation

The location aware collaborative filtering (LACF) is one of the most successful technique of predicting the Quality of Service (QoS) in Internet of Things (IoT) service recommendation systems. However, the openness of CF web service recommendation renders them vulnerable to the injection of attack profiles consisting of apocryphal QoS values (also identified as shilling attacks). Combined with location factors, such profiles might exert greater impact on the LACF compared with traditional CF method. Unfortunately, to the best of the authors' knowledge, there is few research on such kind of attack model in the literature. Therefore, in this paper, the authors first construct three kinds of attack models including LAA, LAB, and LAR (location aware - average, bandwagon, and random) models and compare the impact of the classical shilling attacks (CSA) and location aware shilling attacks (LASA) on LACF. Furthermore, the authors use two attack detectors to compare the robustness of CSA and LASA. The experimental results on WS-DREAM dataset indicate that the LACF indeed suffers from CSA and LASA. Besides, in comparison with CSA, the LASA models do not always exert more influence on the LACF and the profiles injected by LASA are easier to be detected.