Now the world can detect if a politician owns some fake followers or identify some bogus product reviews listed on a shopping site. All this has become possible through an open source algorithm that called FRAUDAR.
A ground of researchers from Carnegie Mellon University has developed FRAUDAR with an ultimate aim to spot fraudsters on social media. The algorithm is claimed to be “camouflage-resistant” and provides “upper bounds” on the effectiveness of fraudsters using real-world data.
“We propose FRAUDAR, a novel approach for successfully detecting fraudsters under camouflage, and we give provable limits on undetectable fraud. We provide data-dependent limits on the maximum number of edges a group of fraudulent adversaries can have without being detected, on a wide variety of real-world graph,” the researchers wrote in a detailed paper.
FRAUDAR uses limits on undetectable fraud as well as provides novel optimisations for detecting scams in social media updates. Additionally, it has the capabilities to detect fraudulent activities.
“Essentially, the algorithm begins by finding accounts that it can confidently identify as legitimate — accounts that may follow a few random people, those that post only an occasional review and those that otherwise have normal behaviors. This pruning occurs repeatedly and rapidly,” explained Christos Faloutsos, professor of machine learning and computer at Carnegie Mellon, in a press statement.
The researcher team has proposed a novel family of metrics to satisfy axioms. To distinguish fraud and normal behaviour in real-world data, the algorithm comes with an improved theoretical bound through novel optimisations.
One of the prime advantages of being an open source code, the algorithm is scalable. It can even incorporate some complex relations available in various usual contexts such as review text and IP addresses.
The code of FRAUDAR is available for free access on the Carnegie Mellon community site.