Rating algorithm is too naive


Right now the value stored in the Rating field of each package in the feed is the average rating. We need to change it to something that takes the number of votes into account and also number of downloads into account.


davidebbo wrote Apr 26, 2011 at 7:34 AM

Not quite following you. What formula do you suggest to calculate the rating? My take is that the rating value should not take into account those two values, but we can treat a package as unrated until it has been voted at least 5 times (or something).

dfowler wrote Apr 26, 2011 at 7:49 AM

Whatever we do, we need to make it such that a package with 1 rating of 5 doesn't rise to the top.

Haacked wrote Apr 26, 2011 at 3:42 PM

We should require at least 5 ratings (or some such number) and the secondary sort should be Download count, not alphabetical as it appears to be.

sebastienros wrote Apr 26, 2011 at 3:46 PM

Don't call it Rating then, but popularity. http://blog.linkibol.com/2010/05/07/how-to-build-a-popularity-algorithm-you-can-be-proud-of/
Though you still "rate" a package, it's ordered by popularity.

kevink wrote Apr 26, 2011 at 6:20 PM

Yeah, I don't think we want to change the meaning of Rating. I remember seeing that blog post about building a popularity algorithm before - don't remember if it was Sebastien or Phil, but somebody mentioned it early on in the project as a future consideration. Now that we have ratings in place, it would be a good time to look at implementing that algorithm.