Hey! Thanks to the whole Reddit mess, I’ve discovered the fediverse and its increidible wonders and I’m lovin’ it :D
I’ve seen another post about karma, and after reading the comments, I can see there is a strong opinion against it (which I do share). I’d love to hear your opinions, what other method/s would you guys implement? If any ofc
Individual users having some sort of reputation is useful. I always thought it was handy on Reddit to be able to distinguish people I happened to disagree with from actual trolls. The latter always had pretty high negative karma scores, and it was good to know that there was no point in engaging with them.
This is why it’s useful at the account level. It’s also useful at the post level in order to build a sorting algorithm which raises the most engaging/important/interesting submissions to the top. Within a community it is important to help define what that community is - irrelevant and low effort content is suppressed and relevant/high-effort gets boosted. Moderators can enforce this by just removing and pinning too, but that’s almost always too unilateral, and the voting system is generally better because it’s expected that then you get a representation of how people in that community feel about it. It’s a good system.
I can imagine some tweaks to help improve how karma is implemented:
Use Bayesan Inference to produce a ‘shit/shinola score’ for contributors instead simple up/down vote totals
Experiment with different recency biases for the score; you can trust that people will change over time
Generally figure out what you’ll be using karma for and make sure you have a way to measure how well it’s working
I’ve googled Bayesan Interference, however I don’t understand what you meant by it. Could you elaborate please :)
Here is a good general explanation of Bayesian inference.
I think @[email protected] is suggesting using such techniques to predict “troll” or “not troll” given the posting history/removed comments/etc. My personal thought is that whatever system replaces karma, it should be understandable to the typical user. I think its possible Bayesian inference could be used in developing the system, but the end system should be explainable without it.
Thanks for the link. To anyone that does’t know about Bayesian inference, do check it out!
Now I have an existencial crisis thanks to the video 😂 the funny part is that thats the same thing used to detect spam email…
Spam detectors are pretty opaque by their nature. In contrast, karma is pretty easy to understand: “x number of people upvoted comments or posts from this user”. This lets people understand a score even if they don’t agree. If a karma replacement behaved like a spam detector, it would probably just annoy people.
Sporting brackets may be a better analogy. They are developed with statistics in mind but are understandable to the average sports fan. I think a karma replacement should have similar properties.
That sounds awesome! You seem to be quite knowledgeable about mathematics. May I ask what is your background?
Computer science. However, statistics is more of a hobby than anything. I am just intrigued by the idea of federated social media in general so I have thought a bit on how I would personally make it work. Perhaps I will make some more in depth blog posts about my ideas at some point.