Katkat
Back

Identify Metrics to Detect Fake-Account Activity on Facebook

MetaData Scientist · Onsite
MediumData ScientistOnsiteStatistics & Math

Last updated

This interview question evaluates statistical assumptions, formulas, estimation strategy, uncertainty, edge cases, and interpretation in a realistic interview setting. A strong answer for Identify Metrics to Detect Fake-Account Activity on Facebook states assumptions, handles edge cases, explains trade-offs, and shows how to validate the result clearly.

Scenario

Facebook wants to understand and curb fake-account activity over the past month and at sign-up time.

Question

List product or engagement metrics that could serve as proxies for month-over-month changes in fake-account activity. Which user-behavior signals would help distinguish fake from real accounts post-registration? Can we flag a fake account during the registration flow? Propose features and quick checks. Given a prior fake-account rate of 2% and a binomial observation of 18 flags in 600 new accounts, compute the posterior probability the next account is fake (clearly show Bayesian steps). To evaluate a fake-account classifier, would you optimize precision or recall? Discuss pros, cons, and the wider impact on Facebook.

Hints

Frame the Bayesian update, assume Beta-Binomial conjugacy, and relate precision/recall trade-offs to user experience and integrity.