Facebook is turning to machine learning to help protect its users from financial scams and the fake accounts that spread them, saying that it took action against more than 500,000 accounts in March.
Product manager Scott Dickens explained in a note that the social network is using previously confirmed scams to help train its machine learning models, saying that it looks for “instances where people are reaching out to others far beyond their typical network of connections, or in unusually large volumes,” as well as similar behavior patterns.
Its not all on the machines, though: Dickens said Facebook has “a dedicated set of trained reviewers focused on taking down scams,” and it also relies on reports from its community.
And he described how suspicious accounts are dealt with: “When our systems assess that an account is likely associated with scam behavior, the account owner must complete a few actions to demonstrate that they are not operating a fake account or misrepresenting themselves. Until they do this, the account can’t be used to reach others. If the owner fails the verification, or if our Community Operations team determines that there is a violation of our policies, the account will be removed.”
Dickens said users who spot suspicious posts or messages should report them to Facebook, adding that the social network will notify law enforcement directly when appropriate.