Would this be a better way of showing statistics?

[DANDAN_THE_DANDAN]

This is just a random shower thought I came up with but wouldn't the statistics represent a more faithful representation if it's highly personalized? This is what I mean.

For this example, let's say there are three mangas: A, B, and C.
A is mecha,
B is romcom,
C is isekai.
And for the sake of the argument, let's say that the true quality of the mangas are as follows:
A - 8.5/10
B - 9.1/10
C - 5.2/10

Let's say that there are 100 people who serve as raters for all three of the mangas.

Each person's tastes will be accounted for with a "taste variable". For example, Bob is really into romcom and isekai but hates mecha with a passion. Because of this, Bob rates the following:
A - 1/10
B - 9/10
C - 7/10

An algorithm will judge Bob's tastes from his readlist, or better yet, Bob outright tells the system his tastes. Then it will give a "taste variable" to each of his ratings. This way Bob's bad rating for manga A will be paid less attention upon when counting the average. Let's say it will give it a weight of 0.05 out of 1 (when in reality the weight is decided by the algorithm).

This way we can filter out bias of dislike of genre.

And now let's say John's a newbie and rates the generic isekai manga C a 10/10. The algorithm takes this into account and gives his rating a weight of 0.3 out of 1.

This way we can filter out bias of inexperience.

And then Matt is an avid fan of isekai and is well aware of its terrible writing. He rates manga C a 5/10. The algorithm is aware that Matt knows what he's saying and gives his score a weight of 0.9 out of 1.

This way we can promote people who are experts of the genre.

And if course, the system will be able to tell troll 1/10 apart from the real 1/10.

If we want to get even deeper into this, we could also make it so that the algorithm gives a personal rating based on the reader's tastes as well. This way, even if the true rating of manga B is 9.1/10, a person who doesn't like romcom will see the rating of 5/10 or something closely related to their feelings of the manga. Basically a way to predict if the person will like the manga or not.

What do you think?

 

[sterven]

I'm not liking this idea at all to be honest. Why should we give more weight to someone's word (rating/review) simply due to their current interest?
What if their taste changes in the future does the score go down or does it stay the same?
The rating system is fine as it is in my opinion so please don't try to fix something that isn't broken.

 

[DANDAN_THE_DANDAN]

@sterven The algorithm changes the weight constantly according to the user's tastes. That means a rating left a long time ago will have various impacts depending on that user.

If the user left a 1/10 a long time ago and happen to suddenly like the genre some time in the future, it's pretty likely for the user to return to said manga and update their ratings. In cases where this does not happen, the algorithm should probably have a system to take this factor into account as well.

I'm not trying to replace MD's rating system, heck, this miracle algorithm is probably something only Google can make. I'm just asking if this idea has any solid ground at all.

 

[sterven]

@DANDAN_THE_DANDAN
Help me get a clearer picture will ya. I'ma throw you a scenario just so I can figure out how this would work please bear with me since i'm slow.

Spoiler

I very rarely read shoujo so rare in fact that i'd be hard pressed to even find 5 series on my list let alone name you any. However i've recently come across this series and i quite enjoyed it. Gave it a rating of 8/10. I must also mention that this series is also an isekai which I tend to avoid (now).

How would my rating be affected overall according to this algorithm and since I avoid these two genres would the initial displayed rating be abysmal or would it reflect the 'true' rating?
How about after I rate it?
Because if the initial displayed rating is bad due to my tastes it would most likely discourage me from reading this series which I would otherwise quite enjoy.

 

[DANDAN_THE_DANDAN]

@sterven

About your concern for the algorithm to censor potentially-enjoyable series:
If it's me personally, I would place the algorithm's score right next to the average score since the function of the algorithm is just to attempt to weed out biased reviews to get closer to the "true" score. And with all algorithms, it may not be perfect thus why the traditional average score will be right next to it so the users can make their own judgement. It's not purposefully trying to censor scores and series that you may potentially enjoy so you don't need to worry about that; the last paragraph above "What do you think?" is probably way too advanced to implement since it carries way too much risk of just being wrong. So just forget about that bit.

Regarding the bit with your rating affecting the series as a "newbie" to shoujo and isekai:
I don't know. The purpose of a neural network algorithm is to tinker with maths to reveal something a human can't.

The perfect algorithm would be able to tell that, even for someone who doesn't usually like isekai or shoujo, if said person can be seen to rate typically well-written series with good scores then there's a good chance that their ratings for this manga is not biased and thus it will give the user's rating a decent weight.

Of course, this is probably something locked behind years of development for the algorithm to become perfect so...

 

[Holo]

There is never a "correct" way of giving ratings - your idea would place a greater burden on the server when calculating stats. The question is whether the extra load is worth the better accuracy (or perceived accuracy)

 

[DANDAN_THE_DANDAN]

@Holo Yeah I can imagine that lol. Nobody is rich in servers enough to be capable of actually bothering with this.

 

[schlo]

@DANDAN_THE_DANDAN
i can see a few problems: I admit it I like Isekai trash. I'm thinking about going to a help group for this. But in the mean time my ratings are probably going to be biased and I'm probably going to rate isekai higher.

"This way we can filter out bias of dislike of genre." How many are doing this? I dislike mecha enough that in many cases it is a game stopper for me. If I read one and think it's great maybe that one should be ranked higher since it might be an indication of "truly good" story that would be loved by mecha fans, who are already likely to read it, and non-mecha fans who might otherwise ignore it. To me many of the things that contribute to a story being a good story are often independent of genre, for example characterization, consistency, plot, art.

Why do you think people's list is representative of what they like? My list isn't complete, far from it. Some of it is what I like a lot, some is just based on frequency of release. Why do you think a large number of people would necessarily go back later to adjust their score. I don't regularly update chapter read, usually it's just spur of the moment update for me. (just an example of updating something)

Maybe they do this already but to me it would be better to have the following criteria to have your ratings counted in the stats at all First x number of ratings (x=10, 20...) before you ratings will be used . The point of this is simply to support the second criteria of user average ratings. If your rating after x number of ratings average is 1 or 9 there is something going on. It might be as simple as you like just about anything or are never happy with anything. It might be that you are playing games with a particular manga or author. That is what I would use.

 

[Alexandregd]

@DANDAN_THE_DANDAN I'd recommend you look up https://en.m.wikipedia.org/wiki/Recommender_system and https://en.m.wikipedia.org/wiki/Utility if you want to learn about the subject of ratings and recommendations.

 

[DANDAN_THE_DANDAN]

@schlo Huh, haven't thought about positive bias as a possibility. Yeah I guess the "perfect" algorithm must take that into consideration as well.

And now that you pointed it out, yeah I don't think readlist will suffice the algorithm, I'm no algorithm expert, after all. If it needs more inputs then I'm not really sure what else besides read time per page and manual user opinion input.

@Alexandregd Thanks for the links but I don't want to spend my afternoon in that rabbit hole.

 

[schlo]

@DANDAN_THE_DANDAN

aside from bias, their is also just how do you rate something. Is art equal to story, more or less impotant? If something is really enjoyable, more so then other stories, but not of great quality, how does that get rated and reverse, great quality but miserable read? I'm willing to bet from conversations I've seen elsewhere there will be all kinds of differences.

 

[DANDAN_THE_DANDAN]

@schlo Humans aren't deciding the weights, the algorithm will. I'm just telling an idea where a neural network reduces all sorts of biases by using user data to get as close to the "true" rating as possible

 

[schlo]

@DANDAN_THE_DANDAN ok, I think I overlooked or misunderstood something at the beginning. You mean the rating would be different from user to user. For manga "title x" it might show a 9 to me and a 4 to you correct? If so I'm not interested in that. It sounds like a way to say this will be in your comfort zone or your safe zone.

Please note I'm looking at this from a practical view, what is doable and not just by MD but even by google. Of course my view of google and yours may differ. I find their search results to be very weak. I would much prefer something along the lines of the old card catalogs used by libraries with some of the capabilities and advantages of computers thrown in.

 

[DANDAN_THE_DANDAN]

@schlo Nope, you're still misunderstanding. The ratings will be the same for everyone. The algorithm's job is to consider how seriously it will take a user's rating. This is because it will reduce that rating's weight according to how much bias it is perceiving based on the user's data. Thanks less weight the rating is, the less it will contribute to the "average" rating, which doesn't really make it an average anymore. Just forget about Google and MD cause I'm just talking about this stats as its own thing.

 

[schlo]

@DANDAN_THE_DANDAN ok so my original understanding was correct. The thing I was trying to point out is what is "true"? Once you get rid of bias what is going to be your criteria? What is true? It is true that water freezes at 32 fahrenheit, that is something measurable once you come up with a standard but what is the definition of good in manga, art, movies, food... What differentiates between a 4 vs 5 vs 6? Is art a bias or a standard? Flip the idea around, it sounds like you're coming up with a way to say title x is a 7 and have it be correct. With all bias removed, what makes it a 7, not talking a number based on votes but what will a user base his decision on?

Actually a more basic question. What is your goal in coming up with a "true" rating? How and who will use it? What will it mean to me or someone else? Basically the saying, "One man's truth is another man's lie." comes to mind when it comes to Art and ratings. Once you remove bias how is it useful to me?

The more interesting idea to me would be trying to achieve my misunderstanding. Matching my ratings to something I like or other user ratings that match what I like.

 

[DANDAN_THE_DANDAN]

@schlo Again, no humans know. The neural network algorithm will figure everything out all on its own because the purpose of a neural network is to figure stuffs out that human can't. We feed it all the user and manga data we can and let it figure things out.

 

[schlo]

@DANDAN_THE_DANDAN i get that what you want to do it with a neural network algorithm. I'm probably not doing a job of explaining my question or my thoughts on it. If you don't mind lets start from the end product rather than the process. Ok, a neural net does everything it is supposed to and the manges "tropes are us" is rated a 3.

What does that rating in general mean, not using the word true?
What does that mean to me or how will I make use of it?
Is the intent that it would affect my selection?

From my interpretation of what you have said the NN would still be affected by ratings of art but would be able to remove the bias. If this statement is correct what would it mean to a manga that has minimal artwork like the Veggie Hot Bun (the name still gets me)? Let's say everything about the manga is top notch. So using all the data fed in it comes up as an 8. How would it be used by me, what would it mean to me, who considers art to be number 4 or 5 on the things I care about vs the guy who says art is the most important, more important then story? Basically what does a non-biased rating mean to people who have biases in what they like to read (not talking genre)?

 

[schlo]

sidenote: honestly what I would be more interested in is a system that can look at the genres and themes and first weed them out by applying a standard definition and then somehow rate them by percentage of the genre/theme in the manga.

For example, my definition of comedy is that their is an intent for the mange to be funny in general. I'm not saying other manga won't have some humor, maybe as a break or moment of relief in an otherwise serious story. The comedy doesn't necessarily have to be there in every page or even every issue but the intent is for the manga to be funny.

 

[Teletta]

@DANDAN_THE_DANDAN
You had a very funny idea, and it's clear that you're smart too, but you are really missing something here.
The thing is, the average is useful just to overcome the problems created by the people voting even if they know nothing. Your system would be useful only if the "bad" people voting are a vast majority.
Personally, I don't think that's the case.
Also, even if it was, your idea would be still ethically wrong.
Indeed, it's like if at the elections, the people's vote would have a different weight based on their curricula.
Like this, for example
Non graduate _ 1
High school diploma _2
Bachelors_3
Master_4
Phd_5
A thing like this is know to be wrong in political philosophy, since it accounts social differences, such as intelligence, class and individual experiences and opportunities.
The society is not and has not to be what it is right, but what represents it as a whole.
It should be our goal to make its outcome the best it can be, but not by artificially manipulating it, but trying to improve everyday as a community, slowly, but surely.


I'm not English, I tried my best not to do huge mistakes, but it sure is difficult 😄

 

[schlo]

@vhk, @DANDAN_THE_DANDAN nah, I've seen his posts around here and he's a good guy. He's just trying to figure out what would make ratings a more useful tool for selecting what to read. Which, to me is an interesting and useful idea. Maybe I'm reading more into his comments but I took it as it would be an additional rating, similar to how there is the mean and the bayesian ratings.

As far as the politics example, I don't think it is the same. First voting affects decisions being made that will end up controlling you and there is only one option, vote or don't vote. For something like choosing a manga to read there are many methods, ratings just being one of them and as mentioned above there are already two ratings, mean and bayesian. What if they chose the mode or median to set the average rating instead of the mean?

@vhk, just wanted to add, I found what you typed readable and understandable, I wouldn't worry about the fact that English is not you're first language. It's much more impressive then a schmuck like me who only knows one language.