Update: Normalized scores, vote changing, vote...Creator: : Matt
August 4, 2010 5:55am
Ok, I'm brand new here. This is my first post. Hell, I haven't even downloaded LoL yet. I have however been poking around here for a few days reading up on builds (because I'm that kind of obsessive researcher) in preparation for playing. I'm posting this because I'm in the fairly unique position of having BOTH scoring systems be new to me.
When I started reading the scoring was (best that I could tell) simply a sum of +1 votes and -1 votes, correct? I might be wrong as I feel like I recall seeing at least one build with a score of 12 that was listed as 24 of 24.
Now the so called "normalized" scoring seems to simply be a percentage, or +1votes/totalvotes. Is this correct? This doesn't seem like appropriate weighting to me (really, a 12/12 build should be given greater weight than a 40/50?), and if this isn't the new system I'd like it to be clearly posted.
It's been a while since I took stats classes, but it seems to me that it would be fairly easy to weight the scores by combing the two. Couldn't you just do (positive votes - negative votes)*(positive votes / by total votes)? I feel like something along those lines would give a good weighting to the scores. It would show clearly the difference between a build that is +20 and -5 (weighted score of 12) and one that is +30 and -15 (weighted score of 10). It's essentially multiplying the old score (a net of positive and negative votes) by the current "normalized" score (a percentage of positive votes). It controls exposure level (old scoring) with approval percentage (new scoring).
I don't know, maybe I'm just a L2PNOOB, but that's my solution.
Also, +1/-1 binary rating system is worlds apart from a 5 star system. One only needs to look at the difference in RottenTomatoes.com (binary) and Metacritic.com (10.0 point scale) reviews of movies to see how it alters the scores.
Also on an even greater tangent I find that a 4 star system is more effective than a 5 star system for constructive input. 5 star systems tend to just normalize to a bell curve while a 4 star leads to people actually having to decide on a yay or nay like binary, but allows them to have a choice gradient.
Anyway... yeah, my first but probably not last TLDR post here.
stuff I said
And... now I see that it could be kind of confusing when I use "/" as both a division sign and to signify x number of y. Hopefully it's not too hard to read.
On the Like it stuff i posted Matt.
It would be just an option, not a tool to compare anything. You could not filter it via "show me the build with most Like it tags" It would be a number somewhere that GETS RESETED TO 0 AFTER 30 days. This is what i ment. It was totaly connected with "no votes or scores at all" idea.
After a month the builds with top 3 Like it tags would get a medal for "most liked" build, similar to "most viewed" guide in those 30 days.
So far my Trynda build was fairly new, it got 6 (+1) votes, and it is shown in Top 30 days builds. And by so, it got -1'ed (ofc without ANY reason) just coz it was at 100% and it looks ugly doesnt it? again i have no idea what that 2 ppl saw so bad in this build that they -1'ed my build. I dont know who did it so i could confront with that person thinking.
Also, to normalize the guides scores you cant count that it will balance itself. As i posted, the most viewed, commented, and top ranked builds on old scoring system are 150+ 'ish on the list. You would have to FORCE ppl to -1 all the "new and 100%" rated builds to even the chances.
Also we are thinking too much on "new = good". Why? I know we want to make new builds and new ideas to shine. But is is a good way?
As mr. Shaftoe said, a build with 6/6 100% score is considered better (by like 150 places on the rank ladder) than 999999999/100000000 build. Also, in this normalized scoring it lacks +0 rank for a build imho (for ppl that accualy see the build and are not amazed by it but also cant say it is bad) And in this situation you are either forced to post a +1 (coz it is not bad) and push this build with it's 100%. a +0 is a must imho in a % based voting for "this build is good but needs some work to earn my +1) also this would encourage ppl to accualy post 100% complete guides not "under neverending and neverhappening construction".
PS. The survey lacks answer that i would like to pick "I dont like it but the old one wasn't good either" :)
Also a small glitch, after leaving a vote you will see the old style score of number not a %.
We're considering an algorithm that will weight the scores based on vote counts, similar to what you said Shaftoe but not exactly. We're running some tests to see how it will affect various builds before we apply it to the live data. I'll let you know how it goes. This would resolve any issues with low vote count builds ranking inappropriately.
The 3/4/5 scale rating might be doable, we have discussed this a few times. One thing at a time though :)
The percentage scores have some issues right now but we will sort them out. I really feel that once it has been finessed a bit, it will be much better than the previous system.
"She put a bag on my head. STILL COUNTS!"
Alright our testing is complete. We have put out an update that modifies all score ordering and score displays based on a Bayesian estimate formula. This is the same formula IMDB and many other sites use to rank content based on user voting. We are still tweaking it a bit but we felt it was ready for use. Note that this hasn't actually made any changes to your scores in the database, it only modifies how things appear on the site, so it's not permanent and we can keep tweaking.
The end result is that builds with fewer votes have their scores weighted down. This means that those 3/3, 8/8, 24/24 builds will not necessarily defeat builds with 50, 70, 100+. The more votes a build gets, the higher our confidence in their score becomes, and the less of an affect the filter will have on their score. Just play around on the browse page a bit and you will see. Let us know what you think.
"I can get an iPhone without giving any money to Apple. I'LL BE LIVING THE DREAM."
Wait wait, did someone just say they LIKE one of our score/voting-related efforts? Hold on lemme screen cap for posterity ;)
Heheh I keed I keed.
We will be tweaking the formula some more in the coming days, but we won't be making any drastic changes to it. It starts to get REALLY mean to lower voted builds, to the point where after the first page every score is in the yellow, or worse. Not to mention we didn't just make this formula up, it is tried and true, so we don't want to mangle it too much.
"Like a boss."
Scroll to bottom. Or just Google for "Bayesian ranking" or such. You can't abuse it, sorry ;) And if you try, big brother will know!
"I hate hospitals. And doctors, and lawyers, and cops. You know what I don't hate? I don't hate vests."
weighted rating (WR) = (v ÷ (v+m)) × R + (m ÷ (v+m)) × C
R = average for the movie (mean) = (Rating)
v = number of votes for the movie = (votes)
m = minimum votes required to be listed in the Top 250 (currently 3000)
C = the mean vote across the whole report (currently 6.9)
ok so for our community what are the numbaz? :>