[TYPO3-typo3org] Ratings, another try

[Michael Scharkow]

Hi *,

I hope to finish the rating system for t3.org this week, but I need some 
help and hints.

Okay, here's the problem to solve:

Ratings are split in three dimensions and relate to *one specific* 
extension version. For every extension version it is fairly easy to 
compute the average overall rating (overall is computed from the three 
single dimensions and stored in a dedicated table column). No problem 
here. All averages are only displayed if there at least 3 (or 5) ratings.

However, this means that on any extension version update, *all* prior 
ratings are "lost": Let's say tt_news 2.2.25 is the best rated 
extension, Rupi uploads a new version, and now it is not even in the top 
list.

Solution 1: Average ratings are always for *all* extension versions, 
this is as easily computed in SQL, but poses a new problem: The top 
rated extension might not have one single rating for the current 
version, or a long-time bad-ass extension might never make it above 
average because of the old ratings legacy, even if the current versions 
are top rated.

Solution 2: Make the ratings for the current version weigh more than the 
older ratings. I don't want to write a research paper on this, so why 
not simply say

WEIGHTED_AVG(version) = 2*AVG(version) + AVG(olderversions)) / 3

So that new versions matter, but the rating history is still useful. The 
problem is now on how to compute this for the long lists of extensions 
(think full list!). Please don't come up with a

foreach(extVersion) {compute_avg();}

because we need this to be *fast*. I don't know if it's possible to 
compute this straight in SQL, so my solution would be another caching 
table in which extkey, version, weighted_avg_rating are stored and which 
is updated when a new rating is submitted.

With both solutions the problem remains that it is possible that a 
overall top rated extension does not have any (good) ratings attached to 
the current stable version, which is displayed everywhere...

Any thoughts on this? And please, no ideas without implementation proposals!

Cheers,
Michael