Hacker News new | past | comments | ask | show | jobs | submit login

"...An Amazon spokesperson told BuzzFeed News the percentage of inauthentic reviews on the platform is “tiny,”..."

This is clearly total and utter crap, as anybody who has spent any time on Amazon knows.

What was not mentioned in the article, but is critically important, is that many times you can get more mileage out paying for negative reviews of competitors than you can your own positive reviews. Many times I've been looking at product X only to read in the reviews something like "X was okay...and enjoyed it. But it wasn't near as good as Y"

And over to Y I'm headed. It's an in-site referral.

I've read many negative reviews where it's obvious that the competitors are paying. "Great look, but why can't they have A?" (A is not important, but another vendor offers it and wants to use it as a market differentiator)

I don't see any easy answers to Amazon's problems. As they continue to monopolize all commerce, it'll just get worse.




> This is clearly total and utter crap, as anybody who has spent any time on Amazon knows.

"Today's Deals" is absolutely infested with brand new products with tons of suspicious reviews, and has been for the last year. Even just checking the date from the first to the last review (e.g. one week) reveals problems for a product with 100+ reviews.

> I don't see any easy answers to Amazon's problems.

There's no "easy" answers, but user reputation is a good start. Right now every review is worth the same as every other review, but looking at the data available we can rate reviews:

- Did the verified buyer actually receive the item before review?

- Is the reviewer's account brand new?

- What does account activity look like outside of reviewing (e.g. Do they browse products? Shop? Check order history? How many non-buy/non-review interactions do we have for every review posted?)

- How did they get back to the product to review it (e.g. direct link? third party site? order history?)

- Did they use a coupon for a discount on the product? Is that coupon posted on Amazon? What % was the coupon? If it is 90% are they even a verified buyer?

- Was the shipment tracked? If not how do we know they even received anything?


I've heard of folks gaming at least some of these points by offering refunds (through a different channel, e.g. paypal) to buyers to leave a positive review. I don't know if it's true, but it definitely seems plausible and an easy way to buy reviews from 'legit' buyers.

Amazon could counter this by offering a big bounty on turning in sellers that do this, and/or financially penalizing them (the sellers) if sufficient evidence is produced.


>I don't see any easy answers to Amazon's problems.

Machine learning is the obvious hot answer. If you could train a model on how accounts that tend to post bad accounts behave (like another reply said, do they browse a lot? do they tend to hit every product they order via a direct link?), you can set up a decent filter. Even things humans might not necessarily notice, like what time of day they tend to write their reviews or their zip code; AMZN has a treasure trove of data to work with.

Given the other story about AMZN on the front page currently (the book publisher's account being suspended), AMZN probably isn't afraid to let the AI ban accounts straight out and let them apply for reinstatement.

But then again, I also saw something today (sorry can't remember where) that the Data Scientists are all 100% engaged in building models that make people buy more, not fixing things. Because honestly, does this problem make that many people avoid AMZN completely? No.


I used to sell iPhone apps (GeeTasks) and I had fans raving about my products, and sometimes trashing competitors. It was weird. Most certainly I did not pay for it.

What I learned from that is that there are people very different from me, with a different way of creating and expressing relationships, loyalty, and adversity.


It’s not just “fake” versus “authentic”, but the incentives are tied to star ratings, and so the text and the rating can be completely opposed. I recently read a review for a refrigerator that (no joke) read “It’s loud, and fails to maintain proper temperature. 5 stars.” WTF?


Perhaps that was a response to a "give it a 5 star review and you'll get a $100/50/10/5 refund"?


I suspect so. I too have revived such a solicitation, and then left a review point out that they were paying for 5 star reviews.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: