
Qualified = count(opinions) > 7
compscore = sum(avg(nonNULL(opinion.category))/count(avg(nonNULL(opinion.category)) except Overall and Re-view
Note: The top and bottom 10% scores are removed from the average calculation (also known as the "Olympic Average")
score = Bayesian Average(avg(compscore+overall+review))
Bayesian Average = (v ÷ (v + m)) × R + (m ÷ (v + m)) × C
where:
R = average for the video (mean) = (old score)
v = number of votes for the video = (votes)
m = minimum votes required to be listed in the top 10% (7)
C = the mean score across all videos (average all vids)
LenWidleheyt wrote:Does anybody know what this means? Could someone explain it with words instead of formulas?
How do you get into the to 10%
Ashyukun wrote:Put simply, as I understand it (and somewhat explained above)- the highest and lowest scores are dropped, and then the final score is weighted such that the more opinions you have for a given average score, the higher ranked the video will be. So, for example, if a video just hit the 7 opinions requirement and happened to end up with an average the same as say, Doki's "Right Now" video, which has several hundred opinions- "Right Now" is going to be decidedly higher ranked.
Scintilla wrote:Not strictly higher, but further from the global average and closer to the video's actual raw score. It's just that all the videos on the Top 10 list have higher raw scores than the global average (else they wouldn't be there in the first place), so THEIR scores are all getting higher through the Bayesian average.
I've been told that the Bayesian average system actually gives lower-ranked videos on the Star Scale <i>lower</i> scores the more star ratings they get.
me wrote:dwchang explained it pretty well in the site announcements thread
here's what i gathered:
Bayesian Average = (v/(v+m)) x R + (m/(v+m)) x C
(v/(v+m)): increases -> 1 as you get more ops. Example: 8 ops = .53, 100 ops= .93, 500= .99
R: This is your "score," what would have been used under the old system
(m/(v+m)): decreases -> 0 as you get more ops. Example 8 ops=.47, 100 ops=.066, 500 ops=.013
C: This is [the average score of all vids on the site] around 8, it won't change much.
Your Bayesian average is a sum of two parts.
(v/(v+m)) x R can be thought as the "certainty of your score." Your video becomes more "certain" with more ops.
(m/(v+m)) x C can be thought as the "uncertainty of your score." Your video becomes less "uncertain" with more ops.
As you get more ops, (v/(v+m)) increases and (m/(v+m)) decreases, so the bayesian average shifts towards R.
As you get a higher score, R increases, so the Bayesian average shifts towards a higher number.
me also wrote:i have a simple demonstration of the bayesian average.
let's assume the average score across all videos is 8 (i donno what it exactly is, but we'll just say 8 for now). under the bayesian average scoring you can think of every video out there starting out with 7 opinions each with a rating of 8. The first "real opinion" counts as the 8th opinion to get factored into the score. so you get 7 opinions with a score of 8, and 1 real opinion of whatever score the reviewer gave you. The second "real opinion" counts as the 9th opinion, and so on.
It's like in some classes the teacher tells you that you're starting with a "100" and you have to do well on tests and homework to keep that high average. well, in this case you start with an 8, and move from there depending on your real scores.
the purpose of the bayesian average is purely statistical. pushing more weight on to the real scores of videos with more real opinions. it's similar to how we reason. if someone told you all of the cats in her town were white, you'd be skeptical if you saw one or two white cats, but more convinced if you saw 100 white cats. similarly, if someone told you a video was good, you'd be skeptical if it had one or two ratings of 10, but more convinced if it had 100 ratings of 10.
x = (8:100); %Votes variable
C = 8; %Site score average
R = 5; %Video score average
m = 7; %Minimum number of votes
for i = 1:93,
a(i) = ((x(i)/(x(i)+m)) * R);
b(i) = ((m/(x(i)+m)) * C);
end
plot(x, a, 'rx', x, b, 'gx', x, a + b, 'kx');
axis([8,100,0,10]);
xlabel('Votes');
ylabel('Score');
title(['C = ', num2str(C), ', R = ', num2str(R), ', m = ', num2str(m)]);
legend('Video Average Component ((x/(x+m)) * R)', 'Site Average Component ((m/(x+m)) * C)',...
'Total Bayesian Average ((x/(x+m)) * R) + ((m/(x+m)) * C)');Users browsing this forum: No registered users and 1 guest