Silk Road forums
Discussion => Silk Road discussion => Topic started by: Mister Dank on November 29, 2011, 07:52 am

The idea that one score on a small order outweighs a good score on a larger order is sketchy. Can anyone explain, in detail, how feedback is scored? I don't want a general answer, I want numbers.

Hello again, Mister Dank.
There was a thread with the info in it back in July.
Clicky on the 'Search' link (not the search box, but the link under your name and stuff at the left top corner of the page.
in the 'Search for' box enter 'feedback'
in the 'by user' box enter 'Silk Road'
check the 'Search in topic subjects only' option
Clicky on the 'Search' button
Clicky on the thread title titled 'Seller ranking and feedback overhaul'
Skip about 1/2 way down page 3, where SR explains:
we can definitely share the detail, though it gets a bit technical, and we are tuning some the parameters, so the whole thing isn't set in stone.
basically, each feedback score is weighted by three factors: buyer activity, age, and purchase price.
The buyer activity weight is determined by finding the average and standard deviation of the log of the number of vendors reviewed by buyers who've ever left a review. if the log of the number of vendors reviewed by the buyer who left the feedback is greater than the average, the review has a "buyer activity" weight of the number of standard deviations above the average plus one. If it is below average, then the weight is the inverse of the number of standard deviations below the average minus one.
The age weigh is found by raising 1.02337 to the power of the review's age in days and then inverting it. We got 1.02337 by making a review half as valuable after one month as one that's just been posted.
The price weight comes from the following formula: 0.2*log(0.02*($price)+1)+1 where $price is in dollars.
You then multiply the weights together to get the total weight for that review. Add up all of the weighted reviews and divide by the total weight to get the weighted average, then do 25*(1avg) to get "percent positive reviews" metric.
To get the seller rank, we find the averages and standard deviations of the "percent positive feedback" and the log of the total weight of the reviews for all active sellers. We then average the number of standard deviations a seller has above or below the average with a 4 to 1 weight on their average feedback over their total weight. This is a seller's rating, and we rank everyone according to this rating.
There are a few little details left out for brevity, but that's basically it.
Also, on the first page of that thread, he states:
Our old feedback system, as great as it was, was overly simple and had started reaching it's limits lately. Inspired by many of your suggestions for improvement, we've done a complete overhaul that I hope you will find more informative and help you determine the best sellers to work with.
Instead of using a simple average of a sellers feedback scores, now a weighted average is done. The weights are determined by several factors. They are, in order of importance:
1) How active the buyer is in rating other sellers
2) How old the rating is
3) How much was spent on the transaction
also, multiple reviews from the same buyer count for less and less. In this way, a seller's feedback score is dominated by recent, unique feedback from active, experienced buyers.
In addition to this, we've developed a ranking system that takes into account not only the average feedback for the seller, but also how experienced they are so you can quickly find the best sellers who have both good feedback and lots of it.
Hope this was what you were looking for.
Cheers,
VJ

This can also be found on the wiki now:
http://dkn255hz262ypmii.onion/wiki/index.php/Feedback_Score
Peace,
DigitalAlch