Adversarial reviews. The problem isn't just fake positive reviews. The other problem is fake negative reviews left by competitors, which is already a problem on Amazon. Most 1-star reviews are either stupid & irrelevant (it arrived a day late, ONE STAR! and so forth), or fake.

I've seen some categories of products with lots of obviously fake reviews claiming safety issues. These were on every single product I could find, so clearly all the manufacturers were targeting each other.

The 2-4 star range has historically been the most reliable, but fake review factories got wise to that a few years ago and will now often intentionally leave a minor gripe in their fake positive reviews to make them seem more authentic.

A better solution is to publish the % return rates for products and forget about reviews. You can even normalise that into a 5-star rating by product category, e.g., "Here are the top 5 products similar to this one, with average return rate of 1%. The product you are viewing has a return rate of 15%".

With low-friction return reason categories you could even classify high return rates for things like "doesn't look like the photos", "poor quality", "smaller than expected", etc. Unfortunately Amazon's return reasons are variable-friction, so their data on this will be biased, although it might be consistently biased across product groups.

I do like the word cloud idea - but pulled from optional free-text return reasons instead of reviews.

This doesn't solve the "it broke after 2 years" problem, but neither does the current review system.

With the amount of money at play, I think return based ranking will pretty quickly result in adversarial return.

The problem is people will do anything for money. Both on the seller side, and on the (fake) buyer/reviewer side.

I recently came across site that's sort of "mechanical turk for scalping", and it was eye opening. IIRC helping to scalp a PS5 would earn 200 in commission.

A vendor with bad ratings in that system could just keep creating new product pages to clean their reputation. There's no incentive to have long-lived product pages.

This is a separate problem Amazon needs to sort out. Far too many duplicate products with a different ASIN. They could easily fix that with manual inventory checking, but they don't. The same process would fix the bait-and-switch problem.

I often search reviews for "linux" to see if anyone mentioned anything like "works out of the box with linux kernel 5.0".

It works the other way too, I don't care if someone gave something 1 star because it failed to work on their windows xp machine.

read 3 star or lower reviews . those tend to be most accurate