Edit: I would have loved to add to the conversation, but people do love their pitchforks and I should be working instead of commenting on HN. :-)
Admitting you worked for Yelp on HN: not even once
Admitting you worked for Yelp on HN: not even once
When cards are cloned, it's exceptionally easy for a bank to determine which merchant is compromised by looking for common denominators.
Take three actors.
What are the odds that two of them have reviewed one business, and only that business? Pretty high. All good. Same for all three.
Now, what are the odds that two of them have only reviewed two businesses, and both businesses are the same, on both accounts? Still decent, sure. All three actors, same two businesses reviewed, and only them? Getting a little dicier.
Two actors, seemingly independent, have each reviewed three businesses only. And they've reviewed the same three businesses, in the same order. And perhaps they each reviewed the first business, then two weeks later, the second business, and then the third. Odds? Getting pretty slim now. A third actor, doing the same? You're stretching the bounds of credulity for sincere reviews.
This stuff is basic probability theory. Looking across reviews for sets and unions of reviews and reviewers, and increasing the number of reviews, but still seeing commonality between them points to SOME malfeasance worthy of review.
I worked on an anti-fraud team in a similar role, and I agree with GP that it's a harder problem than people realize. It's easy to come up with a simple metric that would identify the fraud that you observe, but you don't know what false positives it would catch as collateral damage.
Like this example that you find wildly implausible:
>Two actors, seemingly independent, have each reviewed three businesses only. And they've reviewed the same three businesses, in the same order. And perhaps they each reviewed the first business, then two weeks later, the second business, and then the third. Odds? Getting pretty slim now. A third actor, doing the same? You're stretching the bounds of credulity for sincere reviews.
I think it's highly likely that if you used this metric to find patterns, you'd find fraud, but you'd also find lots of legitimate reviews (e.g., friends who go to the same restaurants together, friends who recommend restaurants to each other).
I don't have the full text of that comment, as they've edited it our, so I don't know specifically what they said. But perhaps they just don't know everything about how Yelp operates; it's not a small company where everyone knows everything.
I also know someone who previously worked at Yelp, and they've talked about some of Yelp's shady internal practices. Everything is not all above-board there.
Gold Plated Bugatti's do not buy themselves, after all.
Is it actually? My bank calls me to ask which transactions were mine and which ones werent, thus requiring outside feedback
What they are looking for in a different context is "Customer A reported unauthorized transactions, and so did Customers B through D. Each of these customers, prior to these reports, all had a transaction occur at XYZ gas station, so is there a skimming problem at that gas station?"
If sales people etc have special access to add or remove reviews, which they do, or at a minimum leave negative reviews themselves when you don't pay, it is corrupt. It is well documented at this point that they do this.