I agree in general, but literally all you have to do is reverse image search new product listings against existing products and then add a call out to manually review this. This isn’t even complicated stuff.
To get fancy you could compare against the millions of known frauds and apply that to new products.
The thing that bugs me is that they do nothing. This stupid “it’s all on you to register your brand and yadda yadda” would be cool if they also coupled that with a ton of effort and smarts to stamp out counterfeiters.
I mean they run mechanical Turk, they could just manually review shit.
They should keep a running log of how they think this is important and all the things they try.
You are missing the point. They are not creating a new product that is a copy of the original, they are registering with Amazon as a supplier of the genuine product, but shipping counterfeits into the warehouse. They are then pricing their counterfeits dynamically so that they always get the “buy box”.
There’s no machine learning to do. The only way to find the counterfeit is to inspect each shipment by hand vs a reference sample to determine if it’s legit.
“The only way to find the counterfeit is to inspect each shipment by hand”
This is basically what the Amazon transparency program is. Is tracks each item from manufacturer all the way through to your door with unique code scanned along the way. You can use the app to see where it came from.
This tells me why companies do not leave decisions on just developers. If we were in a company, someone had developed a machine learning app by now for a problem which needs no ML in the first place.
If an existing seller is listing the product that already exists. It can be examined against other listings and be reviewed. If it’s a brand new seller then interview the seller. It’s not perfect but will provide incremental improvement.
I think the point you’re missing is that the normal course of business for most products is that many distributors compete to sell the same item on Amazon. Go look for example at Canon cameras, you will see 10-15 different legitimate sellers all offering different prices, all on the same detail page - you have to click through to “this item available from xxx sellers”
If you’re interviewing sellers, what do you ask them in the interview? “Are you selling counterfeits?” What are your grounds for allowing them to list an item?
If you want to inspect a new listing, what do you want to inspect? Maybe you inspect the first shipment. Fine. People will figure that out and send a small first shipment of real items, then start sending fakes.
If simple image matching is effective, scammers will just rotate/warp/tint their images. These techniques might also take the legs out from an MT approach. Sellers might also get a product online with one image, then later change to another/reorganize the album.
Perhaps, but that’s harder than you think. Look at all the shit Getty does to find images that have alters, changes, stuff like that.
My point is that they aren’t doing anything. You point out a couple of edge conditions as a reason not to try. Even reducing fakes by 20% would be great.
If this cost Amazon money they would be trying to stop it.
To get fancy you could compare against the millions of known frauds and apply that to new products.
The thing that bugs me is that they do nothing. This stupid “it’s all on you to register your brand and yadda yadda” would be cool if they also coupled that with a ton of effort and smarts to stamp out counterfeiters.
I mean they run mechanical Turk, they could just manually review shit.
They should keep a running log of how they think this is important and all the things they try.