Hacker News Reader

Comment by xiphias2 | Hacker News Reader

What I would like to see is a system which is still binary, but the people that I personally upvote matters.

Just as an example I just upvoted you and I would like to see your future comments, I care less about how many people are upvoting a content and care more about who upvoted it, and whether that person is selective to upvote constructive content only.

View on HN

I've wondered for the longest time about the feasibility/consequences of having a recursive, private karma system (with decay). I upvote you, so I see your stuff, and more of the stuff you upvoted, and less of the stuff you downvoted. You can repeat the process recursively and add decay as you go further. Every user holds their own version of the karma network.

While I read this I tried modeling storage and query models for implementing such decaying recursive karma network. Not quite sure how, it felt hard. If someone knows how to do this, from an abstract perspective or from a systems implementation look, I’d greatly appreciate that! :)

That would be interesting. Might have to add it to the backlog of “cool things to look into implementing”. My main concern is if it would increase the echo-chamberyness of social media.

It definitely would, but I don't feel concerned about it. I think it's an overrated "problem" that misunderstands what people fundamentally want (better ways of building tight communities online).

The main problem I see with echo-chambers today is the culture war type of thing. I think there could be interesting solutions to that: based on the karma "distance", build clusters and adapt network topology so that you can only discover clusters that are at a finite distance. Minimize risks of conflict while letting people explore "fringe" ideas (for them).

Yeah this seems like a way to perpetuate media bubbles.

The website Hubski (a fork of hackernews, but without the tech focus) has an interesting system where you follow people and whenever they upvote something it appears in your feed. I dont know how it would work on a bigger site, since there's few enough posts per day that you can reasonably sift through every single post, but it's a cool concept.

Hey, that's a retweet.

Pretty much, it's a Twitter style feed with HN/Reddit style content (links + comments). https://hubski.com/primer

Hubski looks really cool, except this part:

,,Hubski is temporarily not accepting new users.

Check back soon.''

https://hubski.com/pub/425192

The trollbots and spammers beat us all there.

What you describe is chain-of-trust model and is very slippery slope. Twitter uses similar model (less emphasis on content, more emphasis on who says it). The result is that when a famous person tweets that he sneezed in the morning, that tweet will get thousand likes and probaby get viral. On the other hand, some nameless person tweets that he found astroid that will hit earth in next 10 hours, no one will notice. In my opinion, one the biggest short term application of AI/NLP would be to rank-by-content and moving away from variants of page-rank.

Reply on news.ycombinator.com