Comments on: searching for the Best Stuff on Amazon

Question: searching for the Best Stuff on Amazon

luser — Fri, 02 Nov 2007 13:28:48 -0800

I enjoy finer things, and like to laugh at crappy stuff. Is there any way to search amazon to bring up the best or worst products (in a category or overall)? A list of "all 5-star products" would be too crude, so I'm thinking, search for the combination of higest rating with most reviews, or lowest rating with most reviews.

By: DevilsAdvocate

DevilsAdvocate — Fri, 02 Nov 2007 13:55:39 -0800

I think the mathematical technique you're looking for is the Bayesian average--essentially, it throws in a pre-specified, fixed number of "average" votes in with all the actual votes, so a lone high rating doesn't push something up to the top of the ratings list.

IMDB and BoardGameGeek both do this for their rankings; to my knowledge, Amazon does not.

By: 45moore45

45moore45 — Fri, 02 Nov 2007 15:11:34 -0800

Is Amazon going to admit anything they sell is "crappy" by coding it somehow so you could search it that way? I don't think so. You know who has fun crappy stuff? the Archie McPhee catalog. Rubber chickens and lots of nuns, monkeys in shriner hats, tiki...pure loveable crapola.

By: vegetableagony

vegetableagony — Fri, 02 Nov 2007 16:35:37 -0800

Devilsadvocate is correct about Bayesian Statistics being used for this sort of purpose. The idea is that most things are probably about average, so we can use this "prior knowledge" to temper our predictions about a new item when it only has a small number of measurements (reviews). In practice though for any item with a large number of reviews you will get almost exactly the same value as if you did a traditional average. The difference being that for items with a small number of reviews the Bayseian average will be closer to the overall average of the whole population (ie less extreme, a item with a single 5 star review will be given a rating much smaller than 5).

In essense items with a very high or low Bayseian average will be exactly the items you are looking for.

As to how to search for them, I would hope that Amazon already does this, as many other sites already do

http://blogs.wsj.com/numbersguy/a-closer-look-at-online-rankings-200

but if not you'd have to write a script to do the baysian calculation based on number of reviews and average rating. If you were going to do this I'd suggest looking into brushing up on your statistics first.

Goodluck