Hacker News Reader

Comment by ilamont | Hacker News Reader

I'd like to know how it's possible for third-party sites such as https://www.fakespot.com to more effectively identify fake Amazon reviews while Amazon (with presumably better data) fails so miserably.

Same question applies to Twitter. Regularly we see researchers uncovering evidence of fake accounts and bot networks pushing spam, conspiracy theories, and misinformation (https://www.npr.org/sections/coronavirus-live-updates/2020/0...) while Twitter struggles to identify the fakes.

ETA: Regarding comments along the lines of, "it's in their best interest to let fake reviews continue as it boosts Amazon's sales."

I think this argument is getting a bit old, as Amazon's growing reputation for fake reviews (and fake products) is turning off even mainstream shoppers and giving a huge boost to big & small competitors.

Amazon now has a very big financial incentive to move beyond whack-a-mole and truly tackle these problems, which requires A) doing at least as good as Fakespot at identifying the bad actors and B) implementing technologies and policies that discourage buying or posting fake reviews.

View on HN

Lack of effort? I wrote the FT piece referenced in CNBC's reporting, regarding the 20,000 fake reviews. While the reviews have been taken down, there doesn't seem to have been any repercussions for the companies. All the products are still live... complete with other suspect reviews (just from users not specifically highlighted by our report).

Amazon could do a lot more. I've just opened up Telegram to see what's happening in the scam review groups today, and within seconds I can see that this wifi range extender is being boosted by paid-for positive reviews: https://www.amazon.com/Extender-Wodgreat-Wireless-Repeater-I...

(The 5-star reviews on that page confirm the theory).

If I can do that, on a Sunday morning, from my home... why isn't Amazon?

Seeing how aggressively all the Big Co. optimize for more profits, wouldn't the reason be something like "Doing this at scale isn't worth the effort"

But you can't just remove products for fake reviews. Or else competitors will buy fake reviews for their competitors products in order to get them delisted.

"repercussions" covers a lot if things.

There’s a big jump from committing fraud for your own benefit and committing fraud because it might mess your competitor up. I think you’d deal with a much smaller problem, and, presumably one that companies would be on the lookout for and aligned with Amazon on.

committing fraud is committing fraud. if i'm on a jury, you get no less of a sentence than if you bought the negative reviews.

I’m talking number of cases (how many people relatively would commit the 2 categories of crimes), not necessarily the severity of the crime. But I am now kinda interested in flushing this out lol, Bernie Madoff and Martha Stewart both committed securities fraud, yet served different sentences, but was this ok in your world view?

I don't know the particulars, so hard to say. You're probably wanting me to say intent has bearing. Madoff intended to scam anybody as long as it helped his situation along. Don't know what Stewart's intent was.

You buying positive reviews is intended to make more money for yourself by getting as many sales as possible including if it means another seller doesn't get that sale. The person buying negative reviews is intending to take sales from others so they get the sales instead. Either way, the intent in both situations is to increase one's own position at the expense of others in a fraudulent manner. You're both just as guilty.

For the record, I don’t want you to say anything, I’m just trying to understand where you’re coming from. No agenda with that particular choice, just the first scenario that came to mind of same crime, different sentence.

I don’t understand why you need particulars if fraud is fraud though. They both got difference sentences and did the same underlying crime. Why do we sentence people differently, or is the view that we shouldnt be doing that?

I don't have information on what Stewart did. It's as simple as that. I honestly don't even know what Stewart did. I read up on Madoff. I care so little about what Stewart does, I didn't follow it. Madoff was more interesting to me, so I read more details on it.

Is it a huge jump? Seems to me like it might be a slightly different manifestation of the cobra effect: https://en.m.wikipedia.org/wiki/Cobra_effect

I don't think it is a huge jump. In the past, I was a high volume seller on Amazon and competitors frequently purchased our products and left unwarranted negative reviews or just left a negative review without ever having purchased it. We knew this because in several cases, we matched the shipment address of our shipment with the return address of a competitors product (this was before FBA was huge).

Plus, sellers wash their hands of this by paying a company to boost their listings. It's don't ask, don't tell on how they accomplish that.

> If I can do that, on a Sunday morning, from my home... why isn't Amazon?

Because there's no real way of measuring the cost of a fake review to Amazon. Yes, Amazon could do a lot more. What's the right amount to do? How much should be spent on that initiative, and why shouldn't that money be spent on other initiatives?

Honestly? A lot of effort and money is appropriate. People can and have gotten hurt because of fake, deceptive, or counterfeit goods sold on Amazon.

Safety and genuine goods are important things. If they can't or don't care about that, what makes it any more reputable than random vendors on aliexpress, for instance?

23% 1-star reviews? They need to buy a lot more 5-star reviews for me to even start reading the negative reviews and then to consider to buy the product.

I've wondered the same thing... I've got a few theories:

-Amazon could identify those fake reviews, but doesn't want to play a cat+mouse game and would rather leave the fake reviews as-is, and easier to spot. If they start targeting harder, the fake reviews will get harder and harder to spot.

-Fake reviews (even 5* reviews) may be placed on products by other sellers in a product category to try and get a product banned. The top ranked seller will earn substantially more than the #2 seller. If you're the #2 seller, pay for a bunch of fake reviews on the #1 seller's product, then gather evidence and submit to Amazon to get the #1 seller's product banned.

-Short-term incentive is for Amazon to leave the reviews (higher scored products get more sales). Seems not too probable, but not sure.

> -Short-term incentive is for Amazon to leave the reviews (higher scored products get more sales). Seems not too probable, but not sure.

Amazon has been very fraudster friendly, and without any exaggeration, anti-customer, for a long while. There's several misbehaviors that they tolerate, while at the same time, make it difficult for customers to fight (report, etc.):

- they allow flooding searches with almost-identical products (ebay doesn't)

- they allow replacing a product page with a different one (ebay doesn't)

- they hide shipping prices and actively confuse the customer (ebay doesn't)

- they don't do anything against fakes, even when there are multiple reports (ebay is not great on this, either, though)

There's no reason why Amazon should not (by their strategy, certainly not mine) take advantage of fake reviews.

I worked on this at Amazon a long time ago. One of the goals was to identify and consolidate items that were the same but where sellers intentionally (or accidentally) had split the listing. Amazon has lots of internal metrics showing how correctly matching listings reduces prices for customers and increases sales. Something like 5-7 listings was enough to find a market minimum.

The achilles heal for sellers is that their listings ultimately need to show up When searching for unique information specific to a particular product. They’ll often try to hide that in an unrelated field or in the description. Usually a seller uses the same “trick” for all their listings.

> The achilles heal for sellers is that their listings ultimately need to show up When searching for unique information specific to a particular product. They’ll often try to hide that in an unrelated field or in the description. Usually a seller uses the same “trick” for all their listings.

By unique information, do you mean descriptive adjectives or product features? If so, these terms more often than not show up right in the title:

ProBien Case for iPhone 11, Liquid Silicone Phone Cover w Tempered Glass Screen Protector, Shockproof Durable Protective Rubber Cases Full Body Drop Protection 6.1 for Women Girls-Wine Red

60 Pack Guitar Picks,Cowalkers Abstract Art Random Color Celluloid Guitar Picks Plectrums Unique Guitar Gift For Electric, Bass & Acoustic Guitars Includes 0.46mm/0.71mm/0.96mm

COMFEE' Rice Cooker, Slow Cooker, Steamer, Stewpot, Sauté All in One (12 Digital Cooking Programs) Multi Cooker (5.2Qt ) Large Capacity. 24 Hours Preset & Instant Keep Warm

All of these items were on the first page of results for a single term (iphone 11 case, guitar pick, pressure cooker) so not only is Amazon failing to enforce TOS rules and guidelines against keyword-stuffing, the trick clearly works for sellers.

Amazon and Ebay facilitate fraud on both sides. Amazon, as with Ebay, automates customer complaints to the point of almost no due diligence. There are some very high profile schemes of fraud on the side of buyer where they either claim damages, lost packages, or return empty boxes that force sellers to send replacements or refunds.

Personally I've been able to ask for refunds on Amazon through a 5 minute chat dialogue for products that were months out of the return period (I felt it was in my case because the seller wouldn't honor warranty etc, but this ease of refund/return is abused rampantly on both Amazon and Ebay by buyers).

The first reason is interesting to think about. It's similar to how one needs to be careful in prescribing certain antibiotics, otherwise the bacterial will adopt.

Someone at Amazon had to have looked at the sales boost fake reviews give and decided to leave them. There is just no way that this hasn't been tested. So if the fake reviews sales increase > decrease in user retention, they decided just let the crap fly. This has probably been concluded more than once.

It is too bad. Ironically eBay used to be the place all the random crap ended up, now when I order something off eBay I expect it to show up in better shape than from Amazon. Amazon turned in to the flea market of the web.

> I'd like to know how it's possible for third-party sites such as https://www.fakespot.com to more effectively identify fake Amazon reviews while Amazon (with presumably better data) fails so miserably.

Maybe they can't? Fakespot authoritatively scores reviews but is there evidence that they are accurate? Anecdotally, there are many claims that the results aren't perfect, such as poor scores on a product where the seller knows there are no fake reviews. As a buyer the results on similar, shortlisted, products often seem a bit random compared to my judgement from reading the reviews carefully. Also, without knowing what reviews are identified and removed by amazon themselves we don't have much to compare it to.

I can confirm. My company is a top 2,000 amazon seller (top 0.1%) and we’ve never bought or incentivized a single review. Our fake spot grade is “B” with some products getting “A” and others “D.”

Most recently amazon changed their review gathering algorithm in a way that increases the percent of customers who leave reviews. The acceleration in our review velocity is flagged by fakespot as suspicious.

> "more effectively identify fake Amazon reviews while Amazon (with presumably better data) fails so miserably." ...

Doing so is Fakespot's raison d'etre, but an unimportant side gig for Amazon. I did my MSc thesis on classifying fake reviews (https://douglas-fraser.com/FakeReviews/index.html). It focused on purely text based features, but I am familiar with all of the literature and research at this point.

Non-text based features are more useful (classification based just on text is hard because believable fake reviews are easy to write) but the amount of non-text / behavioral data Amazon gets or would need to process would be a Facebook level of engineering effort - and is it really that cost-effective? (versus letting buyers do the work - humans can't tell if a review is fake from the text very well, but other things like an excessive number of 5* reviews is easy for a human to see)

The removal of these 20,000 reviews (the FT article) is based on a straightforward obvious heuristic and so is cost-effective. But trying to track all the IPs from China leaving reviews? hmmm. But they should have got the distance and banned all the companies associated with the fake reviews; that would definitely send a signal.

Amazon has tried things like Vine etc to deal with the problem; Yelp and TripAdvisor have their own methodologies. But bad actors always seem to find a way around it. Automating the monitoring of communities where spammer groups congregate is possible but that is another major engineering effort - you just can't have humans doing it due to the scale (and Bezos doesn't like paying essential warehouse workers a decent wage either... so why bother with something like this?)

> humans can't tell if a review is fake from the text very well, but other things like an excessive number of 5* reviews is easy for a human to see

Shouldn't an excessive number of 5* reviews also be easy for a machine to see, especially when other indicators are considered?

For most Amazon detail pages I visit, I immediately sort by most recent. I have learned that the most relevant (which tend to correlate with earlier posting dates) are most likely to be fake, sometimes obviously so.

Vine generates real-ish reviews. It doesn't stop fake reviews.

Classic innovator's dilemma. The benefit of spotting fake reviews is too small for a company of Amazon's scale, so the company doesn't spend (meaningful) resources in pursuit of that goal. Amazon has no way of knowing which, among the infinite number of small projects that may end up becoming a commercial necessity, are fated to grow to that size. This allows smaller competition (here, Fakespot) to do move quickly, execute, and seize that benefit which, for a company of Fakespot's size, is highly meaningful. The momentum that a company like Fakespot builds while building out Fakespot's service allows the quality of Fakespot's service to outstrip and embarrass Amazon's parallel attempts.

> I think this argument is getting a bit old, as Amazon's growing reputation for fake reviews (and fake products) is turning off even mainstream shoppers and giving a huge boost to big & small competitors. Amazon now has a very big financial incentive to move beyond whack-a-mole and truly tackle these problems

Citation (unfortunately, still) needed. "Very big" means something quite different to Amazon at Amazon's scale.

They could outsource it. It’s wondrous that they haven’t solved this yet. I have to suspect they prefer it this way.

I am guessing its because the hackers are not gaming fakespot's detection systems.

If amazon adopted fakespot's algorithm, the hackers would circumvent it within a week. This would also happen if fakespot reaches a critical mass of audience.

It Is Difficult to Get Someone to Understand Something When Their Salary Depends Upon Their Not Understanding It

That seems like short-sighted thinking on the part of Amazon though. If I can I am now ordering anything expensive from the vendor directly and avoid Amazon. Just two years ago, I'd have even paid a little extra to have the purchase go through Amazon due to their fast shipping, easy returns and my Amazon store card given me 5% back on purchases. Just the last three months, I probably spent $2k-3k buying directly instead of via Amazon. That was 100% driven by not wanting to get fakes

Absolutely the same for me. Amazon is the new Best Buy for me. Back when online commerce was starting to really take off, it was very common to go to Best Buy to browse then go home and buy it cheaper online. Anymore, I use Amazon for the same thing, I go there to search then go elsewhere to buy.

These days, I only buy from Amazon if I need it quick and it isn't in stock locally.

Why? To save money?

> That seems like short-sighted thinking on the part of Amazon though.

I Amazon makes 1 billion extra each year, it can use it to buy a few of the "vendors" later on.

Profit now has more value than profit in 10-15 years.

Also, many people will be scammed by Amazon/vendor and they will never know it. To detect a fake some times a minimum knowledge is needed.

So, in the very short term, the Amazon customer is happy because he thinks that got a bargain, the seller is happy because the item is not returned, Amazon is happy because keeps their percentage. The authentic brand loses a sale and a year later it even looks bad when the fake fails and the customer blames the original brand.

And, if a customer detects the fake and complains. Amazon/vendor happily apologizes and returns the money. It is just part of being in the fake business.

How does their salary depend on them not understanding it? I would assume Amazon's bottom line would be helped by having employees who fully understand the market impact of fake reviews and can easily identify and take down fake reviews.

When fakespot identifies a fake review there are no repercussions is they are wrong. If Amazon does it, flags or removes those it incorrectly identifies as fake, there will be repercussions with the sellers.

Between all the great AI technologies in development, and hard feedback from decision info that Amazon could get from a fake review removal process, I believe Amazon could easily solve this problem if they wanted.

Bots are the best thing to have happened to Twitter's bottom line. Essentially Twitter at a fundamental level operates by farming the users into nice sorted silos. Then waiting for something to enrage a silo enough for them to obsessively engage with something in a dopamine addiction reaction. Bots are amazingly effective at finding enrage points, and fertilizing what would otherwise be fallow fields. Bots are just a cost effective way of finding the right fertilizer, field, and silo (and all government subsidized on top of it).

> Regularly we see researchers uncovering evidence of fake accounts > while Twitter struggles to identify the fakes.

This is not symmetrical.

This is their job for researchers without doing this they would be unemployed. A month of effort working on a small tight part of Twitter finding a problem and they get a large payday.

Twitter does this everyday and gets no reward other than maybe a slightly better network. They just get people thinking they do nothing.

> Amazon now has a very big financial incentive to move beyond whack-a-mole and truly tackle these problems, which requires A) doing at least as good as Fakespot at identifying the bad actors and B) implementing technologies and policies that discourage buying or posting fake reviews.

What incentive? They're still a giant, and while I've seen people here on HN complain about it, I've yet to see people or companies stop ordering via Amazon due to fake reviews.

Aside from their own sales incentives, why should they invest in something that other businesses (such as Fakespot) are already solving? It's R&D costs they can just not spend at all, and if the public pressure against fake reviews starts to boil up, they can probably buy Fakespot for less than it would cost them to develop the functionality on their own.

Are the HN commenters liars? Many people here swear off Amazon electronics and sometimes books.

The folks here don’t represent the majority. Most people on hn don’t shop at Walmart Or dollar general either but somehow they do alright

What sort of metrics are being used to evaluate fakespot's evaluations? I also think what fakespot does of providing a rating of ratings is a bit easier than actually addressing the larger problem. At some point wouldn't fakespots ratings also be prone to faking as well, surely there will be attacks and attempts at manipulation of that rating system too?

What threshold would be used to remove reviews, how would it be fair to different listings?

Removing fakes can have significant negatives too if you have too many false positives. So to say that amazon or twitter is failing in comparison to fakespot or independent researchers is a kind of an incomplete comparison without knowing fakespot's false positive rate.

In the case of Twitter, they also want to keep the fakes active for a while to help uncover a larger number of accounts at once. Large purges are more effective, because they don’t let the spammers play cat/mouse. Same reason why Google pushes out the large SEO updates in bulk - to stop spammers from gaming them as they come one by one.

What benefit does Amazon get from spotting and weeding out fake reviews?

They gain credibility against rivals such as Best Buy, Walmart or others...

I think that might've been true a decade ago. Now I think it's Best Buy and Walmart that are attempting to gain credibility against Amazon, no?

I usually get electronics from Best Buy as they haven't sent any fakes (to my knowledge...) and it seems that Walmart has a pretty faithful customer base

Tough to generalize based off personal experience but I think overall Amazon would benefit from more credibility and have it be a net positive.

Aren't they all trying to prove that they are more credible than competitors?

I used to buy things by going to amazon, searching for the type of product I want, and then picking one based on user review. For ten years or so this worked very well and I didn't even consider buying things from anyone else.

Now the reviews and rating are meaningless noise and I have to pick out a product to buy by looking elsewhere, which also typically results in me buying things elsewhere.

I agree with you, but my point is that you're the dramatic majority. Their revenue proves it. So, in 2020, even though their reviews are all garbage and spam, does it matter to them? I again ask, now, based on their data, what benefit it poses to them?

This. They are if nothing else very good at data analysis, and when you have achieved market dominance it is more profitable to hide the problem, attack people pointing out the problem, and buy out any one trying to solve the problem than to fix it yourself. They could fill every order today with a box of straw and see no impact at all to their bottom line.

What alternative do people have here in the US? Go to an actual store? I just went to a local Home Depot to buy a power tool. The store was basically stock out on the thing I wanted from every available manufacturer, of which there are 5-6 Home Depot carries. But I could go on the Home Depot website and get any of them. Amazon destroyed physical retail, and now sits on the rubble and collects rent. The supreme irony is that Amazon is now ... opening physical locations. They built their empire on the notion that a storefront is a quaint anachronism of the undigital dark ages, why would they do that?

> could go on the Home Depot website

Not good for Amazon.

I can't edit my comment but replace Majority with Minority and that's what I meant.

It behooves them to solve the problem before Walmart or whoever takes over their customers.

Legit manufacturers hate Amazon due to counterfeits. Ali is cheaper. Shopping websites get easier to make every year. Sellers can hook up to multiple websites easily. What's Amazon's moat? It's been "the reviews".

Walmart can’t seem to figure out search. They are years behind and I doubt their interface will ever catch up unless they outsource

Reply on news.ycombinator.com