Study finds review behavior can expose social links to phishers

Researchers analyzing Yelp data found that an attacker could infer 49% of users' social relationships from review behavior while incorrectly identifying 10% of unconnected pairs as linked. At a 20% false-alarm rate, the study found up to 63% of relationships could be identified. The researchers said correlations in review length were a prominent signal and proposed adding carefully designed noise to released data; simulations found that could sometimes make attack returns negative.