With this many prior discussions here, I guess the only question to ask is: Are any major sites using this method, or at least providing a way to combine sort with a filter items with under X ratings.
If anything, it seems like the likes of Amazon are making it harder to sort and filter, using dark patterns to direct you to their preferred products.
You're right. Remember that Netflix contest where they solicited an algorithm that would make the most accurate movie predictions? It's a quaint vestige from a time before they figured out that the "algorithm" should optimize their profit rather than your media satisfaction, because those are not the same algorithm.
Amazon's product search is largely useless, at least in the UK, now. It typically returns many items that don't include your search term, are outside the price range you specified and are not sorted in the order you asked for. The top half the list being "sponsored" items is just the enshittified cherry on the turd cake.
Also it’s now basically impossible to find the "usual" brands of a product if you don’t know them.
What I mean is, let’s say I want to buy an alarm clock. I don’t want a cheap one neither a luxury one. I just want to buy the medium alarm clock from usual brands you find in store but I don’t know, for a given type of product, which are the usual, reliable brands. Well, Amazon now being Aliexpress with a margin, it’s impossible. Just typing "alarm clock" in the search bar will submerge you with low quality products from temporary Chinese brands.
Yeah, amazon has basically made brands worthless on their site.
I tried searching for a cordless handheld vacuum on amazon recently. All the 5.0 reviews are AI generated unverified purchases. And then just below that, lots of RANDOMWORD generic chinese "brands".
The reputable brands you might actually buy are hidden in the noise of it all.
> Are any major sites using this method, or at least providing a way to combine sort with a filter items with under X ratings
1. Reddit's default sorting algorithm was (and perhaps still is) based on this
2. Google's internal Q&A tool, Dory[1], used it to rank questions to ask to the leadership during the "open" mic sessions.
NOTE: please don't do the same mistake.
The math of the algorithm ensures that controversial questions never get asked, since the algorithm approximated the upvotes/downvotes ratio, not votes/views ratio.
The latter was not used because, IIRC, counting impressions on an internal site was too "technically challenging".
This, in my opinion, defies the entire point of an internal Q&A, which is to address hot topics before they spill out and become an issue.
(I don't think I'm spilling any tea talking about how Q&A in Google was run five years ago; there's no secret sauce there. It was natural to ask "which questions make it to the top", and the answer was the link to the article we're discussing. Google considered it to be fair).
It does. But after a while, reposts are allowed through. That's on purpose, because want good articles to have multiple chances at getting attention.
The reason for posting lists of previous discussions is not to boo reposts—I hope that's clear! It's to point curious readers in the direction of additional interesting comments.
Adding one to both the numerator and the denominator when calculating average ratings isn’t a terrible idea.
In situations where you’re estimating probabilities (like the average rating of an item based on user reviews), there is a Bayesian interpretation of this adjustment.
Beta distribution is a conjugate prior to the binomial distribution, meaning the update process has a posterior distribution is also a Beta distribution.
By adding one to the numerator (the number of positive reviews) and one to the denominator (the total number of reviews plus one), you’re effectively using a Beta(1,1) prior (uniform distribution).
This approach smooths the estimated average, especially for items with a small number of reviews. This is a useful regularisation, pulling extreme values towards the mean and reflecting the uncertainty inherent in limited data.
I'd also include some temporal aspect in many cases, a review on Amazon from the day it arrives is far less likely to include useful information than one from several weeks/months later. This would also filter out a lot of the reviews that seem to review the wrong thing (i.e. reviews that are about the seller rather than the item).
Maybe even start including an incentive to re-review items several months on?
> Many people who find something mediocre will not bother to rate it at all; the act of viewing or purchasing something and declining to rate it contains useful information about that item’s quality.
Not really sure, some people have different base likelihoods of rating something. I will rate apps when I see the app rating popup because it's simple and I know how important that app rating is. But the majority of people will just instinctively close it.
If I see an app rating popup, I will rate the app 1 stars for annoying me with popups.
I'm sick of rating things. I can't write an email to a company, without getting a follow-up email to rate my interaction with them. Same for telephone calls. Every time I charge my car on a public charger, the app will ask me to rate my charging experience. If I order a pizza, I get asked to rate it. Just... fuck off. I'll rate/review things if I feel strongly enough either way about it, but just stop pestering me about it.
Same. I have dozens of Play Store reviews that just say "Kept pestering me to rate, so here's your rating. One star."
I've tried to do this on Amazon for companies that attempted to bribe me for a positive review, but sadly Amazon doesn't allow you to mention review bribes in reviews, and they don't get posted.
In some cases, yea probably. But many things I'm asked to rate, the baseline expectation is that the product or service just works like it's supposed to. For the EV charger, my car is either charged or it isn't. If it is, you have delivered your service as advertised. If it isn't, sure I can indicate your charger is broken or on fire or whatever, but you probably already know because of telemetry. If I'm at a restaurant, sure my opinion can be more nuanced.
The incessant rating prompts just come off as... needy and insecure I guess. And annoying.
At least for android apps, it's best to first rate 5 stars, then the lower score in Play store, as those pop-ups usually only redirect to play store for high ratings, to inflate their rank.
I actually had this problem recently and I just sorted by absolute numbers of positive ratings and it worked pretty well. (You could only rate thumbs up/thumbs down). Nice to see a more mathy solution tho
There are situations where sorting by "number of positive ratings" works, but there are many (I'd say most?) where the downsides are impactful enough to make it a non-option.
The issues are creating a first-mover advantage and a feedback loop. This makes the ranking very... stable, but that's usually not desirable.
The first-mover advantage comes because you're using the absolute number, so things that have been around longer will tend to be ranked higher. Even in a system with positive and negative ratings, sorting by `positive - negative` only reduces the problem... not removes it. Something that's been around for a couple years getting 40% negative ratings is still going up by 20% of the total amount which will quickly be a big number to try and catch up to.
That alone is usually disqualifying, but to make it worse usually the ranking system is creating a positive feedback loop because it's what's driving exposure of the options. People will tend to look at / try / buy / etc the top ranked item before the lowest ranked. So you're driving most new ratings towards the existing top-ranked item and continuing to reinforce its place at the top.
You're basically ranking by "the oldest product that doesn't have net-negative ratings". That may be what you want, but usually it isn't.
(I've never seen it, but I guess if you had only _bad_ options, you'd actually get the opposite effect... the least worst product at the top, getting exposure, and being beaten down until a new least worst product takes center stage and your ranking system just oscillates like crazy.)
Where the simple ranking like this could work a little better would be where all the options are known up-front. There's nothing to "catch up" to since everyone started by the same starter pistol. If you have a positive feedback loop though, you are going to end up weighting earlier ratings heaver than later ratings.
The other case I can think of would be where the options you're ranking are few enough (such that position in the ranking doesn't really create much difference in exposure, reducing the positive feedback loop) and you're collecting few enough votes (further reducing it) and the items are fixed up front (removing the first-mover advantage) that none of this would really have a chance to have any outsized impact. I can't see an issue using this on a poll sent to a dozen people asking them which of 5 places they preferred for lunch.
I've used the "correct solution" from the article in the past. It worked well enough I didn't have to come back to it.
How not to sort by average rating (2009) - https://news.ycombinator.com/item?id=29200103 - Nov 2021 (82 comments)
How Not to Sort by Average Rating (2009) - https://news.ycombinator.com/item?id=15131611 - Aug 2017 (156 comments)
How Not to Sort by Average Rating (2009) - https://news.ycombinator.com/item?id=9855784 - July 2015 (59 comments)
How Not To Sort By Average Rating - https://news.ycombinator.com/item?id=3792627 - April 2012 (153 comments)
How Not To Sort By Average Rating - https://news.ycombinator.com/item?id=1218951 - March 2010 (31 comments)
How Not To Sort By Average Rating - https://news.ycombinator.com/item?id=478632 - Feb 2009 (56 comments)