On the Robustness of Rating Aggregators Against Injection Attacks

For a decade now, Academia has been researching refined techniques to detect fake reviews. In this article, rather than proposing a new detection methodology, we propose to contain the consequences of an attack launched by a fake reviewer who attaches arbitrary scores to the review target. We demonstrate that, by simply changing the score aggregator, the review site can withstands smart and targeted attacks, even carried out for an extended period of time. While experimentation is carried on on real data from a popular e-advice website, our approach is general enough to be applied in any other information service where voting and ratings need to be aggregated.