Power Uploader
- Joined
- Jan 25, 2018
- Messages
- 1,400
So, I've been thinking about corrections to series score inflation.
Bayesian correction arguably works for series that have lots and lots of readers (the idea being to select for series many people like). But it doesn't solve the problem of series that have relatively few readers.
I don't know how hard it would be to code for that, or whether it would be even possible to apply it retroactively (only if the time of grading is recorded by your system).
But how about making more recent user scores count more towards the aggregate score of a series than older ones?
To allow for automated weighting, you can do it like this:
- Scores given by users before 25% of the currently available chapters were out: score x 1
- Scores given by users when more than the 25% and less than 50% of the currently available chapters were out: score x2
- Scores given by users when between 50% and 75% of the currently available chapters were out: score x3
- Scores given by users after 75% of the currently available chapters were out: score x4
- Scores given by users after the series was finished: score x5
Plus, add a correction factor:
- If the user follows the series and has read the latest currently available chapter: weighted score x 1.5
- If the user follows the series and hasn't read the latest currently available chapter: weighted score according to the position of the latest read chapter in the series release timeline
- If the user doesn't follow the series: weighted score x 0.5
And for multilingual users: apply the correction according to the latest chapter in the user-selected language that has the most chapters at the time of scoring.
For series with missing chapters, no way around it, count only the number of chapters in Mangadex. These are a relative minority, though.
That would stimulate users to revise their scores as a given series progresses, and would give less weight to the mob who either gives a Masterpiece or an Awful score based on chapter 1. It would also give less weight to people who drop series relative to those who follow it assiduously.
Of course, this is predicated on the assumption that people who read a series to the end are better judges than those who drop it. In other words, the Solo Levelling crazies would count more than those who were enlightened and dropped it in the middle. Still, I think that for measuring mass attitude towards a series, this is more honest. Aggregate scores are just a measure of current mass tastes, not really a good indicator of narrative quality.
It's probably too hard to code, though.
Bayesian correction arguably works for series that have lots and lots of readers (the idea being to select for series many people like). But it doesn't solve the problem of series that have relatively few readers.
I don't know how hard it would be to code for that, or whether it would be even possible to apply it retroactively (only if the time of grading is recorded by your system).
But how about making more recent user scores count more towards the aggregate score of a series than older ones?
To allow for automated weighting, you can do it like this:
- Scores given by users before 25% of the currently available chapters were out: score x 1
- Scores given by users when more than the 25% and less than 50% of the currently available chapters were out: score x2
- Scores given by users when between 50% and 75% of the currently available chapters were out: score x3
- Scores given by users after 75% of the currently available chapters were out: score x4
- Scores given by users after the series was finished: score x5
Plus, add a correction factor:
- If the user follows the series and has read the latest currently available chapter: weighted score x 1.5
- If the user follows the series and hasn't read the latest currently available chapter: weighted score according to the position of the latest read chapter in the series release timeline
- If the user doesn't follow the series: weighted score x 0.5
And for multilingual users: apply the correction according to the latest chapter in the user-selected language that has the most chapters at the time of scoring.
For series with missing chapters, no way around it, count only the number of chapters in Mangadex. These are a relative minority, though.
That would stimulate users to revise their scores as a given series progresses, and would give less weight to the mob who either gives a Masterpiece or an Awful score based on chapter 1. It would also give less weight to people who drop series relative to those who follow it assiduously.
Of course, this is predicated on the assumption that people who read a series to the end are better judges than those who drop it. In other words, the Solo Levelling crazies would count more than those who were enlightened and dropped it in the middle. Still, I think that for measuring mass attitude towards a series, this is more honest. Aggregate scores are just a measure of current mass tastes, not really a good indicator of narrative quality.
It's probably too hard to code, though.