Forum

Look at this picture:

Now I love the feature that expands the graph. But doesn't something look wrong from weak on down to worst ever.

I am not questioning the math or the numbers(should of cut that part actually), but the whole point of a graph is to give a visual representaion of data. But if you were to just look at the graph itself, you would not realize that more people voted for red than for the second to darkest green.

It also, in this case, would visually show why this is not rated excellent or masterpiece.

I believe the explanation for spam ratings was due to something called a "trust" or "reliability" factor (I forget the actual name, but there was a thread about it). Basically, the Encyc. has a filter that can detect users who keep voting Worst Ever on purpose, or something to that degree.

If the filter picks them up, then it automatically negates their rating, but still keeps tally of their vote.

Or, in plainer words:

The numbers are the actual votes.

The length of the bar reflects how reliable the voters are considered to be, based on their voting patterns. If all you ever do is love or hate, your trustworthiness will be less than someone whose votes are more balanced.

- abunai

Oh Geez!

I generally vote from so-so on up. Would my patterns make my trustworthiness drop below 1? Is there anyway to find out your rating? Or can you give the formula? Does it allow certain people's vote to be worth more than 1?

While I understand why you do this sort of thing, visually it is an afront to my senses. And for lack of a better word, it makes the second part of the graph look retarded. I mean I am a numbers kind of guy, and when I look at the graph it tells me that 77 worst evers ~= 6 goods. Don't take any of that as an insult or anything, that is not my intent. It is just like Data trying to say a contraction.

Dargonxtc wrote:

and when I look at the graph it tells me that 77 worst evers ~= 6 goods. Don't take any of that as an insult or anything, that is not my intent. It is just like Data trying to say a contraction.

I happen to be part of that 6 that voted it as good. I thought that the girl who leapt through time was a good title, it had some issue but still a nice story.

I had no idea that the bar showed how reliable the voters were. So basically the 58 votes for excellent are more reliable then 77 votes for worse ever by 5 times?

Dargonxtc wrote:

Oh Geez!

I generally vote from so-so on up. Would my patterns make my trustworthiness drop below 1? Is there anyway to find out your rating? Or can you give the formula? Does it allow certain people's vote to be worth more than 1?

How could it be more than 1? After all, 100% is 100% is "all there is". And no, there is no way I could give the formula, since I do not know it -- Dan keeps it closely guarded.

I've personally thought about it, and I've come up with some methodologies that would allow me to skew the results, regardless.... but doing so would be a very obvious and dreadfully inelegant hack. Anyone looking for the traces of it would uncover the tampering immediately.

You can't find your rating, it isn't listed. Dan once revealed some of the trustworthiness indices in a thread in the staff forums (unsurprisingly, the staff all had high trustworthiness, being, despite all appearances, sensible and balanced people). But every time you voted, you would affect your credibility, so it would be rather pointless to try to keep track.

Ideally (I suppose) a person with a completely normal distribution (i.e. a perfect bell curve) of votes would score close to 1. That's completely hypothetical.

Dargonxtc wrote:

While I understand why you do this sort of thing, visually it is an afront to my senses. And for lack of a better word, it makes the second part of the graph look retarded. I mean I am a numbers kind of guy, and when I look at the graph it tells me that 77 worst evers ~= 6 goods. Don't take any of that as an insult or anything, that is not my intent. It is just like Data trying to say a contraction.

It's not a an error, it's a contraintuitive display. That's not the same thing. A false display is plain wrong, whereas something can be contraintuitive yet true (ask any physicist).

Perhaps the graph needs a disclaimer to explain that the bar graph doesn't reflect the raw number of votes, but is heuristically skewed to reflect the aggregate credibility of the votes. Then again, most people would just respond with a big fat "Huh?" to that.

- abunai

[quote="Deltakiral"]

Dargonxtc wrote:

I had no idea that the bar showed how reliable the voters were. So basically the 58 votes for excellent are more reliable then 77 votes for worse ever by 5 times?

Note that there are more votes for "worst ever" than for anything short of "masterpiece." The rest of the curve as far as number of votes lays out well enough, but the 77 votes for "worst ever" is clearly the exception. It's likely that the people who voted worst ever go around and rate lots of titles very high or very low whether they've actually seen them or not.

one3rd wrote:

Deltakiral wrote:

I had no idea that the bar showed how reliable the voters were. So basically the 58 votes for excellent are more reliable then 77 votes for worse ever by 5 times?

Note that there are more votes for "worst ever" than for anything short of "masterpiece." The rest of the curve as far as number of votes lays out well enough, but the 77 votes for "worst ever" is clearly the exception.

No, actually the deformation clearly begins at 'not really good'.

abunai wrote:

How could it be more than 1? After all, 100% is 100% is "all there is".

Thats one of the reasons I wanted the formula. When dealing with things other than physics you don't have to deal with absolutes; time, structual integrty, gravity, molecular bonds etc. etc. It is very easy to have over 1, or all there is. Especially when trying to assign a finite number to an abstract idea, like trustworthiness. It's statistics math vs. physics math, and statistics math is the one which would be used here.

But to be fair when I wrote that I was simply thinking along lines of fairness, and not that Dan would logically set the upper limit to one. Embarassed

But that leads me to something else you wrote.

abunai wrote:

Ideally (I suppose) a person with a completely normal distribution (i.e. a perfect bell curve) of votes would score close to 1. That's completely hypothetical.

If it is set up like a statistical bell curve(doesn't have to be perfect), than almost no one would come close to a full vote. Even with a curve skewed toward the limit about 80% of memebers would likely have a score of .4 to .8. Which would mean no one really gives all of there vote. And that you might be considered trustworthy, but your vote doesn't count half as much as member X, who has twice the rating.

Unless there is some sort of outlier cut off when you reach a certain limit. Where say once you reach .2 your vote counts as zero, and everyone above .2 would have there vote count as one.

Interesting. But like you said we really don't know.

abunai wrote:

It's not a an error, it's a contraintuitive display.

I wasn't impling it was wrong. Perhaps you misunderstood my not so obscure Star Trek double entendre. I was simply saying that visually, contraintuitive or not, it is misleading. So I think we were saying the same thing. But you are right, telling people what it is would be an eyesore more than anything else. Oh well.

Anyway thanks for explaining things to me, I hope I didn't bore your head off.

One thing to keep in mind is that people are allowed to vote more than once, for different versions of the show. So they can vote "English Dub, worst ever" "English Sub, Good". Only one vote would however count towards the show.

It also allows people to vote "worst ever" for every version of the show they've seen. Some people try to skew the vote by voting "worst ever" for 30 different languages. But that only counts as 1 vote.

By comparing the total number of people who have seen it (342), to the total number of votes (394), we see that there are at least 52 extra votes, probably more given that not everyone who saw it would have voted.

Those extra 52+ votes could all be from one person, a couple, or 50...

But given the graph, they're probably mostly "worst ever."

Also, as Tony and Abunai mentioned, not every voter is equal. Dan explained this in detail elsewhere,if someone can dig up the post, that would be great.

So even if 77 people voted worst ever (as opposed to 1 person voting 77 times) you could still get the result where the Masterpiece graph would be much longer with even fewer unique votes. In other cases, it might be the inverse.

-t

tempest wrote:

Also, as Tony and Abunai mentioned, not every voter is equal. Dan explained this in detail elsewhere,if someone can dig up the post, that would be great.

This one?

dormcat wrote:

tempest wrote:

Also, as Tony and Abunai mentioned, not every voter is equal. Dan explained this in detail elsewhere,if someone can dig up the post, that would be great.

This one?

Actually, I was thinking of the post he made in the staff forum, oh well Anime smallmouth

;

Tony K. wrote:

I believe the explanation for spam ratings was due to something called a "trust" or "reliability" factor (I forget the actual name, but there was a thread about it). Basically, the Encyc. has a filter that can detect users who keep voting Worst Ever on purpose, or something to that degree.

If the filter picks them up, then it automatically negates their rating, but still keeps tally of their vote.

Does this mean that the filter doesn't look for people who keep voting Masterpiece like spam?

rti9 wrote:

Does this mean that the filter doesn't look for people who keep voting Masterpiece like spam?

No, it doesn't. Just because the police are on a drunk-driving patrol, doesn't mean they won't bust you for jaywalking.

- abunai

rti9 wrote:

Does this mean that the filter doesn't look for people who keep voting Masterpiece like spam?

When user's spam the system, be it masterpiece, the opposite,or something in-between, they're trustworthiness goes down the tube.

Dan and I actually have a long standing disagreement on the best way to handle this.

His method of diminishing the value of a person's votes based on the likelihood that they are spamming the system works quite well, but I feel that this should be augmented with a system that removes the votes of any user, or group of users, that are identified as undoubtedly spamming the system.

To Dan's credit, we've identified many, many accounts that are used for no purpose but spam, and the net effect of these spam accounts is pretty close to zero. So what I want probably wouldn't improve the situation much. I just think of it as an added safe-guard.

-t

tempest wrote:

Dan and I actually have a long standing disagreement on the best way to handle this.

His method of diminishing the value of a person's votes based on the likelihood that they are spamming the system works quite well, but I feel that this should be augmented with a system that removes the votes of any user, or group of users, that are identified as undoubtedly spamming the system.

To Dan's credit, we've identified many, many accounts that are used for no purpose but spam, and the net effect of these spam accounts is pretty close to zero. So what I want probably wouldn't improve the situation much. I just think of it as an added safe-guard.

If Dan's spam-compensation algorithm works, then there is no practical reason to do more.

However, removing spamming accounts (or rather, removing their influence from the Encyclopedia database) may have more than merely practical rating-average related reasons. There is more to an Encyclopedia entry's "status" (for lack of a better term) than merely its average rating, or the sizes of the different bars. No matter how well the heuristics deal with spamming, the number of votes in each bar are still visible, and the result is that some users regard these numbers as more or less authoritative.

For this reason, I agree with you, tempest, rather than with Dan. Obvious spammer accounts should be elided from the database.

If this seems too extreme, then another, intermediate, way of dealing with this issue might be to remove the numerics from the Encyclopedia entries, replacing them with a weighted percentage. That is, stop showing the numbers of votes in each of the rating categories, but show only the bars and perhaps a % of the total weighted votes for each bar.

- abunai

Forum - View topic
Bar graph visual ratios

Taste Naruto's Ichiraku Ramen in Real Life in Silverlake Ramen Campaign

Live-Action One Piece Series Wins at 2024 Writers Guild Awards

Gundam SEED FREEDOM's Rie Tanaka, Sōichiro Hoshi Share Memories at Los Angeles Premiere

Kurayukaba and Kuramerukagari Anime Movies Review

The Spring 2024 Anime Preview Guide

This Week in Games - Ask Yourself, "What Would Shadow Do?"

Forum - View topicBar graph visual ratios

Forum

Forum - View topic
Bar graph visual ratios