Amazon Overhauling Customer Reviews

An anonymous reader writes: Amazon says it's making some big changes to its product review system, one of the most heavily used on the internet and a vital part of Amazon's business. A machine-learning platform will endeavor to select helpful reviews with an emphasis on more recent ones. The average score will change as well: new reviews will be weighted higher than old reviews. Reviews from verified purchasers will have more influence as well, and also reviews voted up by other customers. "For example, sometimes a company will make small tweaks to a product or address some customer complaints, though this product isn't officially updated or renamed. With the new system, [Amazon] said, these small modifications should become more noticeable when shoppers are buying products." Because the review system is so important to customers, Amazon will be rolling out changes slowly, and watching for anything that breaks or gets skewed in unexpected ways.

Comments

Sounds like reasonable changes to me

[FlyHelicopters]

I could be missing something, but frankly everything I read in the summery seems like reasonable changes to me.

Someone who actually is known to have purchased the item, yea, their review should be worth more than random Internet person #4827341

A review from last month is probably worth more than one from two years ago. The product may have changed.

Re:Sounds like reasonable changes to me

[Dutch Gun]

A review by someone who chose a different product is likely to be more valuable in choosing a product, assuming that person can articulate why he/she chose the other product, because that means the person knows not only this product, but also other products on the market. By contrast, someone who chose the product he/she is reviewing has a very high probability of being familiar with only that product and not any others on the market. Given a choice, I'd take reviews from non-owners over reviews by owners any day.

That's some pretty convoluted logic there, at least by my reckoning. If the user hasn't purchased the item in question, how exactly are you assuming he/she knows the product sufficiently that they're in a suitable position to review it, judging it's strengths and weaknesses? Can you assume they're not simply biased against a completing product they happen to own, for example? You're telling me that, for example, a Playstation 4 owner is in a better position to review an Xbox one, and we should trust their judgment as being more objective and fair than someone who actually purchased an Xbox one?

Also, if the product sucks, assuming the product isn't so bad that folks return it, people who own the product are more likely to feel the need to give it better reviews to justify the money they spent.

Wow... again. I'd bet that people who have purchased a product and are unhappy with it are actually *more* likely to review it harshly in an effort to punish the company for their poor product, and at least warn others against a crappy purchase. There are some old marketing saws that say similar things, I believe. At the very least, that holds true for me. I've purchased a couple of stinkers, and I made damn sure to leave a one or two star review, and explain in detail *why* it was such a terrible product.

Heh, your expectations of human nature runs about 180 degrees opposite of mine, for whatever reason.

Re:Sounds like reasonable changes to me

[vikingpower]

Disagree in the case of books. An old review is absolutely not worth less than a recent review, especially if e.g. a classical engineering or math text is 2 decades old.

Stop grouping revisions

[magusxxx]

The other problem is that they'll group different models on the same page. So the reviews you see are for all the models together. In many cases Version B is way better than Version A. But, you still see the bad reviews without realizing it's unrelated to what you're going to buy. They also need to address the paid bad reviewers. I looked up stuff just last week and the same one star review, word for word, was listed on three different items. And these were three totally unrelated items. One in electronics, one in cookware, and the third in camping equipment.

Re:Another company trashes its most useful product

[magusxxx]

Well, yes and no... http://www.amazon.com/gp/custo... ... This person's one star review was posted three years ago and has garnered 76 responses. The most current was 20 days ago. It's amazing seeing the "You're right" - "No, you're wrong" conversation. Even when the company, right from the beginning, stated, "This is how your testing methods are faulty...", people are still saying this one star review has convinced them not to buy the product. [BTW: 88% of 4,121 people gave this product a 4-5 star review. ] So, while this review has survived the test of time, the only usefulness it has achieved is to show how bad testing methods are readily accepted if it's buried in enough data.

The worst reviews on Amazon

[MillionthMonkey]

"I have to give this book one star because I ordered it and it never arrived on time even though Amazon said it left the facility six days before it was supposed to get here!"

"This book is typical LIBTARD crap and if you buy it you're a stupid egghead."

"I haven't read a book in five years so when this book came out I decided to buy it. This isn't the book I thought I was ordering, this is crap written by a different guy with a similar name! Buyer beware!"

Is it really that hard to get a computer to pick these out?

Re:Caveat Emptor

[Anonymous Coward]

You must have missed the bit that says: "Reviews from verified purchasers will have more influence as well".

Similarly, if they're applying machine learning to the whole system, they could fairly quickly pick out people who seem to do nothing but review items on Amazon, or people who create accounts and do nothing but review items, etc.

Gaming the system will still happen, but it'll be harder.

Re:Transcript of a recent meeting at Dice HQ

[AmiMoJo]

Yup. Seriously, was putting the share button where the "read more..." link used to be designed to make people accidentally click on it? Do you guys do any kind of user testing at all, or were you fully aware that's the most clicked on link on the site and just wanted to abuse it?