Stars raw vs. stars adjusted???

Scandia · Post by **Scandia** » Wed Mar 16, 2005 7:50 am

Pardon my ignorance, but....

What is the difference between stars raw and adjusted?

Which stars are added or removed in order to adjust them? Is raw simply the average of the stars you have been given? Are adjustments based on the value/contributions of a given member (video expertise, awards received, donations)?

I could not find any documents explaining this. Please help. Thank you.

Org Profile · Post by **Phade** » Wed Mar 16, 2005 8:44 am

Hey Scandia,

Star Raw is the raw average of the star ratings given by people who are not ignored. Star Adjusted is the Bayesian curved star average (see weekly Star Scale lists). I guess I should come up with better terms for them. How about "Star Avg" and "Star Score"? Is that better than "raw" and "adj"?

I hope this helps.

Phade.

Scandia · Post by **Scandia** » Wed Mar 16, 2005 10:31 pm

Thank you for your explanation, Phade-sama.

I even looked up Bayesian Curve, since I never learned that concept in math. But the "stars adjusted" has helped all my videos.

The "averages" and "scores" terms do sound a bit more self-explanatory.

slackergirl · Post by **slackergirl** » Tue Apr 26, 2005 10:46 pm

Ok, I hate to sound stupid, but I looked up Bayesian curve, and I'm still confused. Near as I can figure, it allows one to predict the outcome (in this case star average) with very few data points. So the more stars given, the closer the average and score values.

It would seem the purpose would be to level the playing field, and predict what a vid's star average would be once it got a lot of stars (say, over 1000).

From what I see, it tends to pull high raw scores down, and poor ones up. I am still totally in the dark as to how or why it does this. Any "Bayesian Curve for Dummies" resources out there? It's been 6 years since I thought about higher math, and I'm a bit rusty...

Scintilla · Post by **Scintilla** » Tue Apr 26, 2005 10:50 pm

slackergirl wrote:From what I see, it tends to pull high raw scores down, and poor ones up. I am still totally in the dark as to how or why it does this. Any "Bayesian Curve for Dummies" resources out there? It's been 6 years since I thought about higher math, and I'm a bit rusty...

Actually, from what I've seen, it pulls high raw scores down and poor ones down too.

No video's weighted star score is ever higher than its raw score.

I believe the method involves taking the raw score and subtracting a certain fraction of it based on how many star ratings the video has received compared to the average number of ratings.
The more ratings a video has received, the closer its weighted score will be to its raw score.

Oslt.

slackergirl · Post by **slackergirl** » Tue Apr 26, 2005 11:07 pm

No, it can pull scores up. Go to "Top Star Scales". If you make the maximum stars a lower number, and go to one of the last pages, the "scores" values are higher then "raw" for lower scoring vids.

Scintilla · Post by **Scintilla** » Tue Apr 26, 2005 11:40 pm

... Oh. Wow. How did I never notice that before?

Okay, then I guess the video's weighted score varies on a scale between its raw score and the global average score, getting closer to the raw score as more ratings are given.

But as to the why of it, I believe the reason was to make it harder to get onto the top list -- you can no longer have a video break out onto the first page as soon as it hits 5 ratings. Thus the top end of the list now more accurately represents what <i>the</i> top videos are.

If there was some reasoning behind the way this affects the rest of the list besides the top, I'm too lazy to think about it right now.

Arigatomina · Post by **Arigatomina** » Wed Apr 27, 2005 12:23 am

Scintilla wrote:If there was some reasoning behind the way this affects the rest of the list besides the top, I'm too lazy to think about it right now.

The only effect on the rest of the list is to make new videos obsolete. And we've gotten around that problem with the option to 'rank vid' according to the true star score, or by the number of stars (or both).

A video with a 'medium' score will get a higher 'weighed score' if it has a lot of stars. That same video would have been near the last page back when it first came out, even if it started at a much higher score - it's the few stars that make the score unimportant. It's the same concept with the top 10 - the more reviews, the higher the score. Only in this case, it's stars. It's useful because you don't have the first few pages full of vids with only 5 stars, vids that would probably fall back on their own if we were patient enough to let them.

Instead of getting the limelight and then falling 'into place' as they get more attention, they start at the bottom and work their way up. This way it's much harder for a video to get to the top, so the list is a better representation of the 'very best.'

At least, that's what I understood from the threads that popped up back when they changed the way vids are ranked. None of the engineering or math courses I took in school used or even addressed the Bayesian method. If they had, who knows, we might have been graded on how many assignments we turned in, instead of how good each assignment scored.

Org Profile · Post by **Phade** » Wed Apr 27, 2005 11:20 am

Hey slackergirl,

Here is my explanation post on how the Bayesian curve works to adjust scores on the Top 10% list:

http://www.animemusicvideos.org/phpBB/v ... 498#515498

The same theory applies to the star scale adjusted score, just replace "old_score" with "star average". Basically, it assumes that the video is an average video until proven otherwise. As more and more scores are collected, the weight of the score is changed from being "average" to being the actual value of the star score. Hopefully the post above will help explain it.

Thanks again and have a great day!! ^_^

Phade.

slackergirl · Post by **slackergirl** » Wed Apr 27, 2005 12:50 pm

Thanks! That really helped!