Top 10% List "Score" Modified

dwchang · Post by **dwchang** » Tue May 04, 2004 11:52 am

Otohiko wrote:Oh God! I just checked out that scale!

Awesome, awesome work. Now I really can't see how it could possibly be better. (well, maybe if it was longer considering I already have about 70% of the videos on it.)

In any case, congrats on the excellent work.

Perhaps a Site Suggestion is adding a filter like "Filter by %" and then you could specify the % you want (like 20 for example). It's not like the scores aren't already there (they're all calculated as such) and it'd just require making a larger table and filling it further.

Another solution (which is already there) is to filter by genre. I know it's not entirely what you want, but by doing that, the list does (to a degree) get longer since not all 171 vids are in there. That way, the vids of that genre that almost made it, now make it.

Bote · Post by **Bote** » Tue May 04, 2004 12:36 pm

WEWT! My video is there to 8) . It payed off dewing all those op. exchanges

.

Scintilla · Post by **Scintilla** » Tue May 04, 2004 1:58 pm

Okay, I see that green denotes a video I've DL'ed locally, but what do red and yellow mean?

Zarxrax · Post by **Zarxrax** » Tue May 04, 2004 2:31 pm

The key is at the bottom of the page.

mckeed · Post by **mckeed** » Tue May 04, 2004 5:13 pm

Ok, I don't understand how this works. I'm math challenged. I had a video that has a average of 9.58 with 16 reviews, and on the top 10% the average is 8.88 or something close to that. How does the average of the video change? I don't get it? Isn't the average the average? What good is the normal average score? I don't know anything about bayes law or why this would be any more accurate than the old way. Are the video averages being scaled? Or do the star ratings have effects on the average now? Ahhhh...my head hurts. Someone please explain, I must be missing something.

AbsoluteDestiny · Post by **AbsoluteDestiny** » Tue May 04, 2004 5:22 pm

"average" and "score" are different things so let's not confuse things here.

The "average" is simply the mean of all the values for each category (review, capture, audio etc). This has always been different from the "score" of a video.

The "old score" gave more weight to 'overall' and 'review'. It would average the technical scores and then add it to overall and review and divide by 3 to find the overall "old score".

The "new score" is worked out by looking at the old scores and scaling them depending on how many opinions a video has. The more opinions, the more "reliable" the "old score" is. This scaled value becomes the "new score".

That's how I understand it at least.

dwchang · Post by **dwchang** » Tue May 04, 2004 5:32 pm

mckeed wrote:Ok, I don't understand how this works. I'm math challenged. I had a video that has a average of 9.58 with 16 reviews, and on the top 10% the average is 8.88 or something close to that. How does the average of the video change? I don't get it? Isn't the average the average? What good is the normal average score? I don't know anything about bayes law or why this would be any more accurate than the old way. Are the video averages being scaled? Or do the star ratings have effects on the average now? Ahhhh...my head hurts. Someone please explain, I must be missing something.

Uhm...phade has the exact formula on his first post. Your average "old" score is just one piece of the new formula. As AbsoluteDestiny pointed out, the new system weighs both your score AND the number of opinions.

Basically the more opinions you have, the more valid your average is. This makes sense since 8 people saying something is a 10 isn't as good (or reliable) as 100+ people saying something is a 9.5. This system takes that into account and I think the results speak for themselves. If you wanna break it down into mathematics, there's a fraction portion prior to your score, this is "how much %" of your average score you get. Obviously the more ops you have, the more % of it you get (i.e. validation of the score).

IMDB.com uses the same system and if you look at their top videos, you'd see a similar trend and agreement (at least to me).

mckeed · Post by **mckeed** » Tue May 04, 2004 5:36 pm

Hmmmmm....interesting. So the more oppinions you get the more of a percent of that average you get. Is there a point where it just says its 100%, like 30 reviews or something, or does it compare it with the number of reviews for other videos and scale it accordingly?

dwchang · Post by **dwchang** » Tue May 04, 2004 5:45 pm

mckeed wrote:Hmmmmm....interesting. So the more oppinions you get the more of a percent of that average you get. Is there a point where it just says its 100%, like 30 reviews or something, or does it compare it with the number of reviews for other videos and scale it accordingly?

Technically there is no point where you get 100% of it, but there are points where it'll probably reach 99% or something.

Again, the actual formula is as Phade said:

Bayesian Average = (v ÷ (v+m)) × R + (m ÷ (v+m)) × C

Now the fraction portion is the first part and v+m will always be greater than v, so the division will never equal 1. Again, when you reach really high numbers like say...200. Since 200/208 is ~96%, that's pretty much the majority of your score.

I don't think it's really an issue to get 100% of your score anyway. Why would it matter when every video is going through the same formula? It would be unfair for some vids to get all of it and others not. This is a dynamic formula that changes based on the score given AND the number of them given.

If you wanna argue that 30 reviews is "enough" to guarantee your score is accurate well then I'm sure plenty of people could argue otherwise. I mean technically a "random video" with 30 good ops could dethrone the established vids with over 200 opinions. To me, that's wrong. A video with that many opinions has obviously proven itself more than a 30 op video right?

That's the reason for such a system. Basically you can't start at the top, you gotta work from the bottom and the more valid your score becomes, the higher possibility of going up. It makes sense imo and in reality, a good video will still be a good video since well...wouldn't a good video continually be validated? This eliminates (the majority) of videos that people would argue about and makes for a fairly consistent and stable list imo.

In theory, with this system, a video will make the top 20 or so very rarely, but let's be honest, how many videos come out in a year that deserve such a spot? One, maybe two? And what are the videos popularing those spots? Surprise surprise, the one or two vids a year that people love. I think that shows the validity of this. Also with this system, the good videos still will make it, but it would take time. I'd rather that be the case than have a list littered with random videos.

Make sense? I know math can suck sometimes

mckeed · Post by **mckeed** » Tue May 04, 2004 5:53 pm

It makes sense, i was just using 30 as a random number I pulled out of my arse.