Amazon Is Rolling Out a Generative AI Feature That Summarizes Product Reviews (apnews.com) 33
Amazon is rolling out a new generative AI feature that summarizes product reviews for customers. From a report: The feature, which the company began testing earlier this year, is designed to help shoppers determine at a glance what other customers said about a product before they spend time reading through individual reviews. It will pick out common themes and summarize them in a short paragraph on the product detail page. The company wrote in a blog post published Monday that the AI-generated reviews are now available to a subset of mobile shoppers in the U.S. across a "broad" selection of products. And it may be expanded to more shoppers and additional categories of products in the "coming months" based on customer feedback, said Vaughn Schermerhorn, Amazon's director of community shopping. The Seattle-based company has been looking for ways to integrate more artificial intelligence into its product offerings as the generative AI race heats up among tech companies.
Fake. (Score:4, Informative)
It's very easy to summarize 99% of amazon reviews.
1 word will do it.
Fake.
A good ML model should spot this and get almost perfect results.
Re:Fake. (Score:4, Interesting)
Well now, let's not be hasty. You make it sound as if Amazon reviews aren't useful at all, and that's not actually true. Amazon review are more subtle than that. This is how to interpret them:
- The product gets a few reivews: it may or may not be good. Double-check outside of Amazon to be sure.
- The product overwhelmingly gets terrible reviews: a competitor can't compete on features alone and bought a bunch of negative reviews. Therefore there's a tiny chance that the product may actually be good. Check outside of Amazon to be sure.
- The product gets glowing reviews: don't waste your time double-checking, it's 100% certain to be utter shit.
Re: (Score:3)
There are other categories:
- This product has lots of great reviews, but the reviews don't actually pertain to the product in the listing. (The product on the listing was "updated," replacing the original product with something unrelated, in order to hijack the good reviews.
- The product has lots of great reviews, but they are too formal (i.e., professional / paid for), or lack detail, or pertain to things not about the product (my product arrived early).
- The product has lots of great reviews, but the sell
Re:Fake. (Score:4, Informative)
My rules of thumb for reading Amazon reviews:
1. Ignore any 5-star reviews. They're either left by the seller's sock-puppet accounts, or by people bribed by the seller to leave 5-star reviews.
2. Ignore any 1-star reviews. They're almost exclusively reviewing something other than the product, such as the delivery time or customer service or some-such, or they're left by people who didn't bother to read the product description before they bought.
3. Pay special attention to reviews with customer photos. Those are the people who are invested enough to give a serious review.
Re: (Score:2)
2. Ignore any 1-star reviews. They're almost exclusively reviewing something other than the product, such as the delivery time or customer service or some-such, or they're left by people who didn't bother to read the product description before they bought.
This is problematic. If you do that, you'll also miss every "product was DOA" review and almost every "product died after three months" review. Mind you, those don't add much in terms of your understanding of how well the product works when it works, but the quantity of those can be an important red flag. You also miss all the reviews where the product differs from the description so badly that it isn't usable for its intended purpose.
So out of my last 20 reviews, I gave:
Re: (Score:2)
Well now, let's not be hasty. You make it sound as if Amazon reviews aren't useful at all, and that's not actually true. Amazon review are more subtle than that. This is how to interpret them:
- The product gets a few reivews: it may or may not be good. Double-check outside of Amazon to be sure.
- The product overwhelmingly gets terrible reviews: a competitor can't compete on features alone and bought a bunch of negative reviews. Therefore there's a tiny chance that the product may actually be good. Check outside of Amazon to be sure.
- The product gets glowing reviews: don't waste your time double-checking, it's 100% certain to be utter shit.
I find my first step with Amazon is to make sure I'm using the vendors store, not some 3rd party.
Any product worth more than a few quid I wish to purchase I'll typically seek out some professional reviews on it. For cheap stuff (I.E. a powerboard or some cheap sporting equipment) back when I had to go to a store for this crap it never had reviews, it was always caveat emptor and still is. You pays your money and you takes your chances. I barely even bother scanning the reviews these days unless it's for
Re: (Score:2)
I'd say Amazon reviews can be summarized as follows:
People who were paid to leave a review, or are otherwise employed by the seller as shills.
People who had the item for ten minutes and are posting a glowing review because the damn thing hasn't broken yet.
People who are pissed off because the item is complete crap.
Always read the 1 and 2 star reviews. Ignore everything else.
Re:Fake. (Score:5, Insightful)
But there's always a huge number of reviews that are nothing to do with how good the product is.
Item arrived broken? 1 star
Item arrived late? 1 star
It haven't used it yet but it looks good! 5 stars
etc.
The best 1 star review I ever saw was for a long string of pearls. The review was "The one in the photo has a knot in it, mine didn't. Seller told me I had to tie my own knot."
Re: (Score:2)
But there's always a huge number of reviews that are nothing to do with how good the product is.
Item arrived broken?
That's actually an important piece of information. With only rare exceptions, if a lot of people are complaining about a product arriving broken, that's a pretty strong indication that the product is not robust enough to survive real-world use.
Re: (Score:2)
Except the shills promoting a product often leave 5-10 negative reviews on competitor's products as well. It's not just pushing your product up, it's pulling your competitor's product dow
Re: (Score:2)
Not even fake, not even the right product (Score:3)
Too often the review algorithm aggregates reviews of several different products into one set. More than a few reviews that are detailed and sound honest are are for a similar product, or a product from the same vendor, but they are not a review of the product you are considering buying. This worse than useless.
And what are the odds (Score:3)
that is summarizes all the reviews, rather than just the five star reviews on the products Amazon makes a lot from, and the one start ones for everything else?
Re: (Score:3)
that is summarizes all the reviews, rather than just the five star reviews on the products Amazon makes a lot from, and the one start ones for everything else?
The problem with Amazon reviews is not a lack of time to browse and mentally summarize the reviews. The problem is a lack of trust that the reviews accurately reflect the quality of the product. The proposed AI sounds like a more efficient way to process garbage in, garbage out.
And will the AI take as an input whether Fulfillment by Amazon and other Amazon income streams are utilized by the merchant? Amazon doesn't just do things for the benefit of the customer, unless there is additional profit for Amaz
I have a better idea (Score:2)
Train an AI to recognize fake reviews and delete them
Reviews are very close to useless
Re: (Score:2)
Can't be done. You'd need to find some non-fake reviews for the training process.
Re: (Score:2)
This is not about being useful. This is about "Amazon offering some service with AI", regardless of actual usefulness.
A new hammer! (Score:2)
"Ooh, a shiny new hammer! Now if only we had something that vaguely resembled a nail to hit with it..."
I only wish that Amazon could have figured out how to apply blockchain to their comments. I bet that would have *really* made the review-bots sit up and take notice.
Different product (Score:2)
My issue with Amazon isn't so much reviews that are fake, it's more about reviews for a completely different product.
When you see a product with a variation selection, all of the variations share the same reviews, even when they are different products.
Screwed the pooch (Score:3)
There is nothing Amazon can do to its review system that will make me trust it. I now shop on-line elsewhere, much less, and shop brick and mortar like it's 1999. Amazon was pretty amazing up until around 2015 before the Asian invasion.
TFS (Score:2)
that should be easy... (Score:1)
Unique data (Score:4, Interesting)
As with many of the posters above, I was skeptical about this, but I have become less so. Hear me out.
Unlike the general public, Amazon has pretty good internal data on the accounts posting reviews. They haven't shared this with us (outside Verified Purchase indicators which may more may not be trustworthy). But they have it.
If the combine review texts with reliability indicators, this could be a real improvement. And...the bar is risibly low.
Re: (Score:3)
I used to post reviews on Amazon (Score:3)
I stopped, mostly because actual reviews are being lost in a sea of noise - some of which will be fake reviews, but some will be from purchasers wanting to sound big and important on the interwebs.
Re: (Score:2)
Re: (Score:3)
I will post reviews if there are not many or if I have a negative review. Not often though and I do not buy a lot from Amazon besides books.
Re: (Score:2)
I not only stopped about 8 years ago, but I deleted all reviews I had previous posted.
Of negative use (Score:2)
You need to read Amazon reviews in order to find what is really going on. In particular, you need to look at the negative ones. Do they have actual gripes or were they just too stupid to use the product right? Positive reviews are generally useless, not only if they are fake. Hence any king of "summarization" will remove what makes the reviews valuable and its results will be a waste of time.
I start with the 1 star reviews (Score:1)
manipulation (Score:2)