4 ms·

FakeSpot and ReviewMeta seem to do a decent job of identifying obviously fake Amazon reviews. I suspect that Amazon just have an incentives problem - purging t

FakeSpot and ReviewMeta seem to do a decent job of identifying obviously fake Amazon reviews. I suspect that Amazon just have an incentives problem - purging their site of fake five-star reviews would go a long way to restoring trust in their platform, but it would negatively impact sales in the short-term. From what I've heard, Amazon has a very decentralised and data-driven management culture, which is antithetical to the short-term pain/long-term gain implicit in fixing their reviews problem.

mike_hearn 8y ago
How do you know what their FP rate is?
- alexgmcm 8y ago
  But equally - does it matter? I would rather have less, but more accurate, reviews than have all of them but many of them being fake. It seems like a use-case where a high false positive rate would be acceptable. Unlike say, medical diagnosis or something.
  - jsight 8y ago
    The problem is that very few really "good" reviews actually get produced. If you accidentally filter out the really thorough and useful ones, you can end up with "real" reviews that aren't well written and don't really focus on actual product issues. Take, for example, the review of the weights in the article. I would have ignored his review, even if it had been posted. He reviewed Amazon's warehouse rather than the product on hand.
  - mike_hearn 8y ago
    If legit reviewers have their work deleted because of FPs they get annoyed and won't write more reviews, they may even stop using the site entirely.
    - alexgmcm 8y ago
      They could shadowban the reviews - so the reviewer would think it was being displayed. A bit deceitful but maybe better for UX.
    - jdietrich 8y ago
      Deletion is not the only option. They can weight the contribution of reviews to the overall score. They can weight the display order of reviews on the product page. They can weight the display order of products in SERPs. Assigning reviews, reviewers and sellers a "fakeness probability" allows you to do all sorts of subtle, useful stuff that clearly isn't being done right now.
- fredophile 8y ago
  If I want to ignore fake reviews because they don't provide me with useful information is it unreasonable to think I would also find reviews that have the same characteristics as a fake review unhelpful? I don't want to spend all day looking through reviews so I think the false positive rate isn't as big a problem for this application as some others.

mike_hearn 8y ago

How do you know what their FP rate is?

alexgmcm 8y ago
But equally - does it matter? I would rather have less, but more accurate, reviews than have all of them but many of them being fake. It seems like a use-case where a high false positive rate would be acceptable. Unlike say, medical diagnosis or something.
- jsight 8y ago
  The problem is that very few really "good" reviews actually get produced. If you accidentally filter out the really thorough and useful ones, you can end up with "real" reviews that aren't well written and don't really focus on actual product issues. Take, for example, the review of the weights in the article. I would have ignored his review, even if it had been posted. He reviewed Amazon's warehouse rather than the product on hand.
- mike_hearn 8y ago
  If legit reviewers have their work deleted because of FPs they get annoyed and won't write more reviews, they may even stop using the site entirely.
  - alexgmcm 8y ago
    They could shadowban the reviews - so the reviewer would think it was being displayed. A bit deceitful but maybe better for UX.
  - jdietrich 8y ago
    Deletion is not the only option. They can weight the contribution of reviews to the overall score. They can weight the display order of reviews on the product page. They can weight the display order of products in SERPs. Assigning reviews, reviewers and sellers a "fakeness probability" allows you to do all sorts of subtle, useful stuff that clearly isn't being done right now.
fredophile 8y ago
If I want to ignore fake reviews because they don't provide me with useful information is it unreasonable to think I would also find reviews that have the same characteristics as a fake review unhelpful? I don't want to spend all day looking through reviews so I think the false positive rate isn't as big a problem for this application as some others.