Rating [1]

Just convince everyone here that a 10 point system is going to be significantly different from a 5 point system. Maybe then people will take you seriously. The fact that changing to a 10 point system will most likely not affect ratings (as evidenced by NU vs. MU ratings being approximately equal).

And all you had to say was "your "refutations" aren't legitimate concerns though.".

This has been brought up before. Unless you provide extraordinary evidence otherwise, and bring up strong points that have not been brought up before, you're not actually doing anything. Just look at the result of your poll.

The practical problem is getting everyone to do it. That's what's introducing a flaw. How are you going to force everyone to do it? Unless you can, then it's a problem.

Did I accidentally write that in French in my previous post? Oh wait, I don't know French... so I wrote it in plain English. Hmmm...

It matches your tastes in novels. With other people with similar tastes.

That's exactly what Amazon does. They will recommend you products based on previous purchases and what other people who purchased similar items also bought. I understand what you were suggesting.

Do you not use Amazon, perhaps? Then you could've at least checked what I was talking about before saying it's not...

Obviously I didn't literally mean every single user. I'm saying most will.

You're willing to bet your life on this? That doesn't mean anything...

And no, it's actually not the same probability if you're going to be pedantic.
We can check and know for a fact that everything is not perfectly correlated. Probability is zero, zilch, nada.
Probability of world ending tomorrow is negligible, but not 0.

Seeing how you're being so pedantic and not really tackling the actual issues, practicality and lack of function, I really can't think you're being serious...

And you said you had to look up an "algorithm" for Bayesian average, yet you're a statistician?
........ I don't wanna go too deep into this end of attacking you as a person rather than your points, but I call bullshit on that.
A bayesian average is simply adding fake votes for ratings that match whose value is a pre-determined average (which can be taken as the average of all novels of the same genre, for example). It's so simple, and you had to look up a supposed algorithm for this? You'll forgive me for not believing you're a statistician.


All right, that used up all my spare patience. Baton pass!

 

There's no need to get everyone to do it.

Multiply all extant ratings by 2 — or live there as they are in the case that half stars are introduced — people who want change their ratings would go and do that.

I'm not necessarily using the same algorithm that exists?

I'm not a statistician by Profession, the same way I'm not a Scientist though I claim to be one.

I merely meant as someone cognisant of Statistics and thinks statistically.

It is of the same order of magnitude — or it is even less.

The probability of all 60,000+ of NUF's user's ratings being strongly correlated is so small, that I'm willing to bet that the world would end tomorrow, rather than that happening.

Amazon matches based on what other people who also bought item X bought —according to what you described.

What I'm describing checks the correlation between all members of the community.

You do not understand what I'm suggesting, because that is not what I'm suggesting.

If two people — say X and Y — have the strongly correlated ratings,
across novels they both read, then they have similar tastes.

If X rates a novel J that Y has not read 5.0, J would be recommended to Y.

This isn't what Amazon does — and more importantly it is not what you described.

This is not the algorithm I'm going for though? The Bayesian Algorithm I'm trying to design doesn't follow this procedure.

I did say — earlier in this reply — that I am not a Statistician by profession.

I am merely cognisant of it.

The same way I call myself a Scientist because I follow the Scientific method (despite being poor even at High school Science).

I think Statistically — hence I'm a statistician.

 

Well,10 point system is a good idea..but seriously,as long as open voting exists with anonymity,ratings can never be trusted. You just have to accept it.

 

I see no other thing than 10 have more number than 5~
which still have lil different to 5 star~

 

"The probability of all 60,000+ of NUF's user's ratings being strongly correlated is so small"
I don't think you even get what correlation is... maybe not even a proper grasp of the concept of probability.
I highly suspect you're one of those people who don't believe the answer to the Monty Hall problem just because it's not intuitively obvious...
Please apologize to every statistician and scientist in the world right now.

Amazon, based on the stuff you browse and buy, will look for people who did the same, and then recommend what they look at and buy to you. How is this not the same?

... I had a feeling this would happen, so I wasn't going to check this thread anymore.
But then I got baited after someone else made another thread asking about the ratings system.

What you basically said there near the end was "I said Bayesian, but I didn't actually mean Bayesian".
Do you even know what Bayesian Probability is? If you're making your own interpretation on the concept of probability, why are you calling it Bayesian at all?

You're trolling, right? Ok, you got me. I'm actually mad now. Kudos.

 

You're wrong. ¯\_(ツ)_/¯

Nope.

My avatar is Bayes theorem though?

I said the algorithm for Bayesian rating I'm trying to design isn't that.

I know what Bayesian average is. I'm not using an algorithm that adds fake votes.

That's where I said that the algorithm I'm trying to design is different.

If two individuals (X and Y) have a correlation of say 0.9 across novels, then it judges that those two individuals have similar tastes.

I.e if X likes J, then so would You and vice versa.

Amazon doesn't — it looks at people who purchased a product and what other products they purchased.


This tries to match the tastes of the users.

I do know what Correlation is.
I also know Bayes Theorem — check my avatar.

I could reply with ad hominem, but I think I'll desist.

 

If my ratings for Xianxia Novels A,B,C,D,E,F,G,H,I (9 novels)
are 3,3,3,4,4,4,5,5,5
And your ratings are 1,1,1,3,3,3,5,5,5
Our correlation based on this data set would be perfect. Rho = 1.
I'll just let that sit here for you to think about.

Words have definitions for a reason. Efficient communication depends on people having the same expectations for what words mean. Going to deep into semantics isn't practical, but there's a limit to using arbitrary definitions...

Bayes Theorem, that formula, is not Bayesian Probability. Bayesian Probability is also known as Evidential Probability and is not something as simple as one formula. Clearly, just because someone uses a formula for their avatar doesn't mean they really understand it.

I could mathematically derive the Black-Scholes model on the fly at this point in my life. Even then, I can't with great confidence say I understand it fully. Much less just putting a formula as an avatar.

And I'll point out that with all your talk of designing this and that, you've shown nothing of the sort.

And don't even speak of ad hominem to me. Have I skipped rebutting any of your points? Even those that I agree with, I didn't simply ignore, but acknowledge, although I did add that the effects are minimum at best.

You were the one who brought up being a statistician, and then made that ridiculous schpeal about statisticians and scientists. Preposterous.

You're the one skipping points and simplifying issues by saying they're wrong without giving a proper reason.

There's not enough facedesk memes in the world that can describe my emotions with regards to you right now. Why am I still here...

 

Because you're bored? Or maybe because you just don't feel happy if you see something on the NU subforums marked as unread?

 

I understand Bayesian Probability—to an extent at least—Bayes theorem is something I know and grasp intuitively; it's what epistemological rationality is based on.

What does that have to do with anything; I said I'm working on a Bayesian algorithm—which I've put on hiatus for now due to work—I never said I have such an algorithm.

You have used as hominem.

You've been attacking me and saying the equivalent of I'm an ignorant person who's spouting shit they don't know about.

You stated:

1. That I may be the kind of person who rejects the solution to the Monty Hall problem just because it's not intuitive. The implications of that makes it one of the largest insults you can levy at me.
2. That I should apologise to all Statisticians and Scientists.

To even try to argue that you did not resort to ad hominem is dishonest at best and an attempt at gas lighting at worst.

I have never once stooped to your level.

Your entire argument about introducing a ten points system creating problems is objectively wrong; I explained so, and you started some inane ramblings about it introducing bias.

I pick my battles; when someone says an argument that is laughably bad —and obviously so —I ignore them. I told you your arguments for the demerits of a ten pint system aren't legitimate —they're not.

This would be my last post explaining why there are no demerits of a ten point system.

1. All extant ratings are multiplied by two to scale to the new system; if someone is d
2. It is not necessary to adjust ratings and those dissatisfied with their current ratings—who would necessarily be supporting the change—can adjust theirs at their convenience.
3. Those satisfied with the current system would he entirely unaffected.

A 10 point system has no negatives—save only for the work in creating it—and some positives; it is the dominant strategy.

You're right, thanks for the pointer.

I've adjusted my system:

Instead of using correlation, we'll have standard deviation.

We have two people "X" and "Y". X and Y both rate "n" novels.

They provide ratings of A_1, A_2, A_3, ..., A_n and B_1, B_2, B_3, ..., B_n respectively.

The mutual standard deviation "dev" of their ratings would be in LaTeX: $$dev = \Sigma_{i=1}^{n} \frac{(A_i - B_i)^2}{n}$$

Now the percentage similarity "sim" of their tastes for X w.r.t Y is given by (in LaTeX): $$sim = \frac{\mu_{Y} - dev}{\mu(Y)}$$


The percentage similarity "sim" of their tastes for Y w.r.t X is given by (in LaTeX): $$sim = \frac{\mu_{X} - dev}{\mu(X)}$$

If the sim is above some threshold say 80%
We can recommend the highly rated novels of X for Y and vice versa.

Alternatively, you could multiply the "sim" by a rating to get a new value K and if K is above some threshold recommend that novel.

I prefer the first method though.

@Nazgand help in presenting LaTeX on these forums?

 

I believe I've made reference to at least one negative that you cannot fundamentally logically refute. A 10 point rating system simply isn't aesthetically pleasing to me and to many others, as can be seen by the poll. Subjectively, your rating system is thus inferior. And while you can talk about bias and logic, the users are the most important thing. You can't have ratings if you don't have users to rate.

 

no
I don't like clutter