Give more weight to ratings from users that leave a wider variety of scores

[TheLunarCarver]

The highly undemocratic idea is that if users leave more scores and make more use of the ten star values, their votes would count more than others towards the overall rating.
Pros:

• Encourages users to leave a wider variety of ratings and think harder about what they put on the top or bottom.
• Gives more weight to users who read a wide variety of manga (like everything from an author, genre, tag, etc) as opposed to just the stuff with the highest rating/follow count.
• Gives less weight to users who just give 10s to anything they like and ignore everything else.
• Gives less weight to users who have just recently started reading on this site and likely have less of a grasp on what they think is good, bad, and mid.

Cons:

• People might try to game the system by giving random ratings to manga they haven't read in order to bomb/boost whatever they rate as 1/10.

Unknowns/Implementation:

• What shape would an ideal user's grades take on? I imagine it would be some kind of bell shape, but the center and spread of said curve probably varies from person to person.
• How much should a user's vote weight increase? I imagine the formula would be something like "1 + (some_constant * log(1 + (number_of_votes * spread_multiplier)))", with spread_multiplier being some value between 0 and 1.
• I can see this being hard for the servers to keep track of. It's probably not a good idea to recalculate a manga's rating every single time a user's vote weight changes from rating something else. Some kind of daily refresh would probably be better.

Side idea I'm less confident in: Also weigh a rating based on how many chapters of the manga the user read.
Pros:

• Makes it harder to review bomb/boost stuff.
• Gives more weight to readers who have more chapters available in their language.

Cons:

• People could just get around this by clicking "mark all on page as read".
• Gives less weight to low scores on any manga that is genuinely so bad that people don't want to finish it.
• Gives less weight to people who already read the manga before finding it on MangaDex.
• Can't be applied to manga with no chapters available.

 

[Manko-sensei]

MangaDex uses Bayesian average as it has been explained in others post such as this one. I don't know where you got the idea that one score has more weight than others, all titles start at 8 and change according to the ratings they're given.

 

[TheLunarCarver]

It sounds like you read this post as my theory of how MangaDex currently works. Unless I messed it up somehow, it's meant to be an idea to improve the system with even more math, which is why it's in the Suggestions area. It should be possible to implement this while still doing the Bayesian averaging.

 

[Angry_Panda]

improve the system with even more math

I get the concept, and don't disagree with it (in fact, it's a good idea in the abstract), but 'put more load on the server by making it figure and then track rating distributions for every user and factor those into the title ratings' is not a thing the devs are going to go for, methinks.

 

[UnnamedPlayer]

I don't see why more varied grades are more genuine than always high/low targeted grades. If people only trash some mangas because they're bad, give very high notes to the best ones and forget to grade the "slop" or the difficult to grade, it's fine. It's information that should be taken in account, even if the statistical distributions of grades would be disrupted.

Your proposal try to find a "normal distribution" around the "real grade", by underweighting biaised graded. But we don't know what should be this distribution, it cxould change from a manga to another, etc. We are talking about tastes, something hard to quantify, that can change with time, and so on. It's useless to have a very reliable grading system, it would be noise averaging over noise with or without bias. The current system is enough.

And i feel like i'm RIP AND TEARING statisticians by using statistics concepts in a profane way, but that's fine, statistics people are soo nerdy even i don't think of them as humans.

 

[GipoScribantino]

I don't remember when or who, but one of the admins/site devs once explained a funny thing about ratings (maybe to explain why they added Bayesian averages? I forgot).
Apparently, out of all the people reading on the site only a fairly small percentage actually leave a rating. And out of them, most do so only when the score is not utterly negative, in that case they simply drop it and don't bother anymore. The most common case a score is left at all is when a series end, meaning most "ongoing" works receive way less ratings than they could.
That is to say, the actual ratings are already very skewed toward uselessness and making them more complex won't help at all.

 

[TheLunarCarver]

I don't see why more varied grades are more genuine than always high/low targeted grades. If people only trash some mangas because they're bad, give very high notes to the best ones and forget to grade the "slop" or the difficult to grade, it's fine. It's information that should be taken in account, even if the statistical distributions of grades would be disrupted.

Your proposal try to find a "normal distribution" around the "real grade", by underweighting biaised graded. But we don't know what should be this distribution, it cxould change from a manga to another, etc. We are talking about tastes, something hard to quantify, that can change with time, and so on. It's useless to have a very reliable grading system, it would be noise averaging over noise with or without bias. The current system is enough.

And i feel like i'm RIP AND TEARING statisticians by using statistics concepts in a profane way, but that's fine, statistics people are soo nerdy even i don't think of them as humans.

I am not a statistics person. That's why my statistics ideas are not good . The hope is that stat robots would see the idea and maybe it sparks better ideas in their brains (even if it's totally unrelated to this one).

 

[panzerkampfwagyu]

I mean, the idea that ratings of users that have been more active on the site getting valued more isn't that abstract, sadpanda does it, for example, both in their comment rating system, but also with their moderation power.
I don't really think that "people spamming random values in order to get higher rating value" is also not likely to be a problem, since most don't put that much effort in to alter their score like that.

Would say something like an account having a rating value in the profile, combined out of things like account age, forum activity, non 10 or 1 ratings, etc. and using that would add too much server load,

 

[jecroisenlasuperioritedubel]

Methinks : We need a kind of Ranked-choice-rating system — 1st preſs to be the 1st choice and so forth.