Hacker News Reader

Comment by mmanfrin | Hacker News Reader

A couple months ago, I googled 'whiteboard amazon', clicked the first result, and was taken to a well structured amazon page for a whiteboard that had 5 stars. Looking at it a little closer, I noticed that of the 143 reviews, 143 were 5 star reviews. On top of it, every review followed a similar structure, was made by an account that had thousands of 5 star reviews, and was so beyond the pale obvious fraudulent that I felt the need to email jeff@amazon.

Today I was looking at some vitamins, and I checked every single result for a certain supplement that was 4 or 5 stars and every single one of them ranked D or worse on FakeSpot.

How the fuck does Amazon not know how to deal with this? 100% of reviews coming in for a product on the same day? MAYBE THAT'S A SIGN, AMAZON.

Amazon reviews are worse than garbage now.

View on HN

> How the fuck does Amazon not know how to deal with this?

How do you know they don't, and have via whatever mechanism decided that they make more money if they choose not to?

It's impossible that giving customers sub-optimal recommendations is good for amazon.

My theory is that they have metrics for selection they are trying to optimize. If customers only bought the 10 best of any item, how could some PM get a promotion for increasing phone charger selection 100x?

Definitely not, long-term. But plug it into a product manager's A/B test and I guarantee it boosts things short-term.

What's to boost? The customer would have selected the better option and been more satisfied, and Amazon would have brokered the deal either way. They still need their hammer.

Maybe in A/B tests, customers who see a product with 4.9 (fake) stars are more likely to click Buy than customers who see the same product with 4.3 (honest) stars?

"We at Amazon think it would be a shame if an algorithm flagged your {insert fake 5-star Chinese product here}.

Have we mentioned we offer a Premium Seller Protection service that will ensure nothing... untoward happens to your listings?"

They are trying and failing - my wife is a seller and easily 20-30% of completely valid positive reviews from repeat customers disappear without a trace.. at the same time a racist tyrade of a review from a buyer who has swastika as user avatar is impossible to get rid of ( 5 or 6 tries so far)

reading this, feels like ebay is a really nice place to sell stuff.

They might have their own issues, but feels more authentic.

Really? I feel selling stuff on ebay has a 1/5 chance get scammed.

I don’t know what categories you shop there, but I have never been scammed. Often make sure to contact the seller beforehand though.

It’s easy to get scammed SELLING on eBay. Buying is not much if a problem as you can get a refund easily as a buyer.

I've sold a couple hundred of items on eBay and I've never been scammed.

I sell a huge amount of stuff on eBay (retro computing / gaming items) and have found it to be a really fantastic and friendly community. Perhaps it depends on the items being sold?

Probably. You are selling in a niche that scammers likely don't care about. If you were selling a Macbook or iPhone, there'd probably be a lot of shady buyers.

I've bought a ton of stuff on Ebay, but only ever sold a laptop when I wasn't getting any bites on Craigslist. It was a limited-edition Dell laptop, so the buyer was really enthusiastic about it and you could tell he was a normal person and not a scammer.

Fortunately I live in a big city now and have never had issues selling in person via CL/Kijiji.

File an online criminal complaint at German police. We really don't like swastikas.

I was in the market for headphones. I bought a pair and was looking for another because I was dissatisfied. I saw my exact model under another listing...but oddly it had ~150 5 stars. I thought no way in hell THESE were loved by everyone. Every single review was made in the same day, with broken English reviews.

These were electronics, and knockoff airpods to boot. But if I 100% need the real thing, I buy else where. For example I just purchased several bottles of Vitamin D for my infant. I'm not trusting Amazon even though Amazon would have shipped faster.

I never buy anything on Amazon that I put in my body. They really need to get this issue under control. It's a huge trust problem for me.

Totally agree. I won't but stuff on Amazon for ME to eat. Much more so anything that's going into a baby's body.

Been using Swanson Vitamins for about 10 years. Decent website, excellent vitamins. Smaller mom and pop b-to-c that has been online since the first .com bubble.

My wife uses it for our family as well. If you watch you can catch really good sales as well.

I find it unlikely, since they don't really have much customer lock-in, so reputation matters a lot. It doesn't make many disappointing purchases for customers to start looking elsewhere first.

There are many 3rd party sites (like fakespot.com) which do a better job than Amazon does and are not incredibly sophisticated. Amazon just doesn't care, that is the only conclusion that is possible to draw from the available data. They are not even doing a half-assed job, we're talking about a fractionally-assed job here, perhaps even milli-asses.

Amazon could easily do at least as good a job as fakespot, and then followup with consequences for review fraud. They could also implement various policies and use the huge amount of data they have available to go well beyond that level and make some headway against the problem. But they haven't done so. Because they can't be bothered. Because people being suckered into buying shitty products hasn't hurt their bottom line yet. They are complicit.

Just because those 3rd party sites are not targeted by fake reviewers.

They are not 3rd party review sites, they are 3rd party sites for algorithmically analyzing the reviews on amazon for a given product.

How would fakespot be targeted by fake reviewers?

Adversarial attacks. You tweak your fake reviews until fakespot is convinced your reviews aren’t fake.

Like not post them at once.

I felt like they know there's a large industry built around it and now that it will likely just adapt to taking simple measures like that. By leaving it the way it is for now it's super easy to spot product pages with fake reviews by anyone that cares to look.

It gets even worse that true 5 stars and low ratings can get flagged for false review by Amazon's algo.

Last year there was a posting on hackernews how this person lost his Amazon account as it got flagged on its first review with a long history of purchases.

> How the fuck does Amazon not know how to deal with this? 100% of reviews coming in for a product on the same day? MAYBE THAT'S A SIGN, AMAZON.

Maybe that's bad for business if they did. People see a < 5 star review and may not purchase it, even if it is a lie.

I think that food supplements/vitamins are the worst. My one and only email to jeff@amazon.com was on that. It probably didn't do any good, but I felt better.

Luckily, they don't appear to be of any value anyway.

Surely their advanced machine learning would see such an obvious pattern?

There are more fake reviews than real ones.

Due to this inversion, it is practically intractable for Amazon to detect fake reviews in a fully automated way. They must, at some point, be human moderated.

The class imbalance might make the modeling slightly more difficult but certainly not 'practically intractable.'

Is it so difficult?

Amazon already has the number of helpful vs unhelpful votes for each review.

They could start with that voting data, and then use metadata (geoip, account usage/posting timing, client fingerprinting) to create quality metrics for each user. They could even have separate trustworthiness ratings per product category (a good book reviewer might not be so good at rating appliances). From there, they'd have a rough idea of which reviews are real and fake, and they could start training a NN on the text.

Figuring out when sellers are complicit, and when it's competitor sabotage, is more difficult, but Amazon has a ton of data about sellers and reviewers and buyers and the social/commercial graph connecting them. Amazon doesn't appear to be working on solving simpler problems, such as incorrect product/book comingling. Therefore, I have trouble giving them the benefit of the doubt that they're working on hard problems like addressing fraudulent ratings/sellers.

Good plan. See my previous comment; I did my MSc on classifying fake reviews. The behavioral / non-text features are far more useful than the text, based on all the research. There is no technical reason Amazon could not capture all the signals, except for the scale of the effort. So I'd say someone has decided it is not worth the effort/cost given the likelihood it won't improve sales that much.

As for NNs, .... an ensemble of classifiers would be best (but of course I'd say that).

"More fake than real"

Source?

Why Google detects low quality sites in its search algo's.

True, but you take some very strong signals to classify a review as (most likely) fake de facto. Such signals could be "reviewed without purchase", "user has over 1,000 reviews", etc.

Reviewed without purchase is a feature not a bug, according to amazon. People don't buy stuff that's not reviewed so they want to catch people who may have gotten it elsewhere.

It's a sign amazon is not playing in good faith.

I did a MSc in ML and my dissertation was on classification of fake reviews:

https://douglas-fraser.com/FakeReviews/index.html

Some years ago, there were some studies trying to estimate the percentage of fake reviews. It came out to 2% to 6%. I'm sure things have increased since then, it does look quite bad given all the anecdotes I have come across.

But unless Amazon has thrown away the bulk of the 140 million or so reviews they have between 1995-2014, I doubt there really are more fake reviews than real ones now, in total.

It would be interesting to do a follow up study to estimate the percentage, given all the blatant issues (such as reviews for other products, etc) Amazon is obviously ignoring. Given what I know, it shouldn't be that hard to process the obviously suspect things but the important question is always: how much profit is there for Amazon to do all that work, given it probably won't increase sales/revenue that much? It is not in Bezos' interest.

The easiest thing would be to block off all of China's IP addresses....

In case you didn't notice yet, it seems like the certificate of your website is expired (2019-04-10).

I would not buy anything that is meant to be consumed on Amazon.

These kind of review patterns are easy to spot. I used to read through recent review etc. These days when I find a product on Amazon that I feel like buying, I go to reviewmeta.com and check the review trend before deciding.

> How the fuck does Amazon not know how to deal with this? 100% of reviews coming in for a product on the same day? MAYBE THAT'S A SIGN, AMAZON.

Then they would stagger the days. The issue is that human adversaries are formidable and you start approaching generalized AI to defeat them.

The issue the tech community has is that Amazon doesn't even appear to be doing the bare minimum.

It's either that, or they are failing really miserably at controlling the flood of bogus/fraudulent review.

I find the latter hard to believe. We're not talking about a mom-and-pop shop here. This is Amazon, they employ many thousands of very smart people that could no doubt build solutions that are better than the status quo.

Absolutely. And note that in other areas (such as payment fraud) they put in a lot more legwork because the consequences actually hit them directly instead of just their customers.

You're right that there is an arms race between fake reviewers and Amazon, but the conclusion - that because it's a race Amazon shouldn't try, is spurious. Amazon should pursue the race aggressively, and because they can allocate more money and software developers to the contest than any one group of fake reviewers, they should also ultimately win.

You're right. Is Amazon not applying any machine learning to their reviews? Google solved spam email and that seems similar.

Google beat email spam, but has fared much worse with fighting blackhat SEO. Email spam has evolved remarkably little in the face of spam filters, and it may be that they were able to "solve" it simply because there never was all that much money to be made from it in the first place.

I don't think there's an arms race. Is there really evidence Amazon has even thrown any shots across any bows?

Per seller stickers on items in inventory so fakes in comingling can be singled out. Bad sellers perma-banned.

Those would be two things you could easily start with, and Amazon knows it.

Allow users to report bait and switch listings and ban repeat offenders.

Require a $1-5K USD deposit to sell on Amazon with a clear fine structure.

I was with you up until the deposit. What about someone who is just starting their own business? $5k is a lot of money

There are other places to start off... also, it doesn't have to be $5K, just cost prohibitive to lose if a knockoff company keeps getting closed out on.

There are also other selling options when starting out... not everyone has to jump onto Amazon.

Can you name one?

I was on ebay, jet.com, walmart, staples, newegg, etc. Anywhere there was a third party marketplace system I was on there.

All of the non amazon marketplaces I sold on combined were a rounding error on my Amazon sales.

I think a deposit is in order, just with the amount tuned.

I think it could be a sliding scale deposit $100 to list a product, and then scale up based on amount of inventory, customer reviews, etc.

If it's a $100 initial deposit that slides up to $1,000 based on volume, that seems pretty reasonable even for something small starting out.

Keep a percentage of gross to go towards a deposit and don't remit payment immediately on new accounts. Imagine a 1K deposit wherein only $200 was paid up front and the rest was paid in terms of 2% of sales.

Further if you spend a week selling counterfeits and as it turns out that you make 800 usd in scam sales. The 2% towards deposits might only be $16 + $200 deposit but if the seller hasn't been paid out yet its trivial to claim the entire $800 in sales as the forfeited deposit.

The seller instead of profiting $400-$600 is now out $200-$400 in inventory and $216 in deposit with nothing to show for it.

The deposit needn't be ruinous. Just enough to make being a bad actor not worth doing.

Not for running a normal business it isn't. A deposit would probably get rid of all the dropshippers, who are really just a form of arbitrage; I think that's probably necessary to get rid of the review chaos. People actually manufacturing their own goods ought to be able to find that kind of money.

However, if we're going to have deposits, there need to be absolutely clear rules on when they are forfeited, with a real, transparent dispute procedure.

  $5k is a lot of money

That's kinda the point of fines, isn't it?

  Per seller stickers

... which themselves would quickly be counterfeited.

Amazon could generate unique per unit tokens for vendors to sticker next to the SKU stickers they already require. Totally traceable to a vendor if there is a problem, and not easy to counterfeit.

This is a beginner programming exercise. They have no excuse for not implementing basic safeguards while they spend billions on AI research.

Amazon applies said vendor stickers when they receive stock from a vendor for inventory.

If Amazon receives product with vendor stickers already attached, they get rejected permanently.

"you start approaching generalized AI to defeat them."

This is, of course, bullshit. Human moderators could defeat them, Amazon is just not willing to hire them. A seller reputation (including bans) could defeat them, Amazon is just not willing to lose the cheap knockoff stores. A listing fee would discourage a lot of the shadier stores, Amazon is just not willing to lose them.

It's not that it's an insurmountable problem, it's an insurmountable problem if you combine it with boundless greed.

We had the same problem a century ago. (Cf. the origins of the term 'snake oil salesman'). We solved the problem, through regulation. It's time we solve it again, through regulation.

I can pretty much guarantee that making Amazon liable for damage from fake reviews will solve the issue. It won't happen in the US - regulation is up for sale here - but it will happen in the EU. It might just be enough to drag Amazon into sanity, kicking and screaming.

I hope it's not enough to drag Amazon into sanity. I hope somebody else comes along and beats the crap out of Amazon by offering what Amazon doesn't - honest reviews and rejection of straight up junk.

And maybe that's the way that Amazon goes down; quality over quantity. Hey, this is HN and we should be able to spot a deficiency in a business plan, right?! Here's how you go about toppling Amazon and building an empire:

1) open a store for online goods 2) take reviews seriously 3) profit!

Let's expand the plans a bit: 1) open a store for online goods 2) take reviews seriously 3) profit, for a while 4) get bought out by amazon or someone else who will then proceed to kill "2" 5) if resisted, be a target of coordinated media smear campaign and/or hacking attempts, each totally not ordered and covertly backed by amazon or another player whose piece of a pie you are eating 6) crack under pressure and just give up

Or in more likely scenario, die away and do a bankcrupt since nobody gives a shit about your crappy market place/online store

Mmm... I was just hoping Amazon's competitors slowly gain an edge over Amazon and that Amazon's reputation is tarnished enough that it greatly impacts business.

> It's time we solve it again, through regulation

You had me up until here. The last entity I would look to to solve this first-world problem is the clumsy, over-bloated arm of government.

If you think this is a first world problem, you might want to look up the history of regulations, and how many people had to die due to shitty products before we finally said "enough is enough".

We're heading that direction again. Our current flavor of capitalism strongly favors externalizing of all costs, and short of enforced liability for products violating standards will always opt for violating whatever low standards bar you set.

(I'd also like to point out that the part of the government that is truly over-bloated is the one that the "lean government" people oddly never want to gut, the military)

If Google can stop email spam then Amazon, who has a reviewer's identity, credit card number, browsing behaviour, and purchase history should be able to spot fake reviews.

That's what I've been thinking. I get more spam than legitimate email, but over the course of a year or so Thunderbird's Bayesian classifier has gotten to probably 98% accuracy with <.1% false positives.

> The issue is that human adversaries are formidable

Are they ever!

Imagine if Amazon had a decent algo that would flag 100% of bad reviews and penalize the seller.

All of a sudden, competitors would be flooding EACH OTHER'S products with fake 5 star reviews to try and put them out of business.

Humans are very crafty when you give them a system of rules. Each twist of the rules creates new opportunities for unintended consequences.

PS: I do not think this is an excuse to give up trying

"and was so beyond the pale obvious fraudulent that I felt the need to email jeff@amazon."

Obviously Jeff is well aware, and could do a lot more about it, but doesn't, ergo he is complicit.

Fake Reviews are 100% supported by Amazon and Bezos.

It's basically full on fraud, right out in the open.

Capitalism dynamics indicates that they DO KNOW that some reviews are fake.

Reply on news.ycombinator.com