enhanced trust algo

We still use a web of trust, but we've refined how trust is established between users. In the previous version, trust was established whenever a user upvoted another user's content. This was convenient and it allowed us to reuse a user's trust to determine their share of daily rewards. No more.

In this release, trust is gained when two users upvote the same piece of content and diminished when one user upvotes something the other user doesn't. A statistical model is used, specifically, a binomial proportion interval, to determine this trust level even when the sample size of upvotes is small.

Why the change? Mostly because if I trust the content you share (by upvoting it), it doesn't mean I trust your upvotes. This release means you only gain upvote trust if and only if you upvote well according to other users.

Everyone's trust has been recomputed and this is live right now. Notably, many more users are now trusted given that they lurk and upvote more than they provide content.

rewards

Given the trust algo change, we needed to change daily rewards too - your upvote trust doesn't entirely account for the value you bring to SN. Also, the way rewards were dispersed was not obviously related to the value you provided recently - if you had shared good content, even in some distant past, you were regularly rewarded. In interviews, most users didn't know why they were getting rewarded.

You now receive rewards strictly depending on the last 36 hours of value you provided. Very good and active users are likely to see 5-20x the daily rewards.

How rewards are dispersed now:

  1. To users who posted the top posts of the last 36 hours, we give 1/3rd of the daily revenue. Users are rewarded based on how well the post ranked relative to the other top posts.
  2. Same as (1) but this 1/3rd of daily revenue goes to users who posted the top comments.
  3. Users who upvoted the content in (1) and (2) are given the last 1/3rd of daily revenue. Users are rewarded based on:
    • how early they upvoted the content relative other users
    • how much they tipped
    • their upvote trust
    • the ranking of the content they upvoted

reward notifications

In a prior release, in response to users saying they were overwhelmed with rewards notifications, we made it so we only showed your rewards once and then disappeared them from your notifications. Reward notifications are back to always being viewable while solving this problem for those users (we identify "islands" of reward notifications and combine them into one notification).

what's planned (barring any fires starting)

  • we are going to improve discussion markdown rendering
    • multiple heading sizes
    • better image display
    • table of contents and linkable headers
    • etc
  • @kr is writing a more detailed FAQ that will take advantage of discussion markdown improvements
  • ~jobs enhancements given both employer and user feedback
  • polls
  • another yet to be determined sub
related

Another thought: The daily reward only rewards the last 36 hours.

That means it doesn't distinguish between users being inactive for a weekend or for weeks. If you look at other social media such as Youtube the algorithm punishes harder the longer people were inactive.

Not saying that's a good or bad thing. Just an observation how the incentivization will work

130 sats \ 3 replies \ @kr 7 Jul

Until now, the daily reward has functioned similar to a dividend… a recurring revenue stream you get for having helped the site grow at some point in the past.

The big issue with it was that people don’t attribute the value they put in with the value they get out. This change tightens that link. Hopefully it also incentivizes people to participate every day, something other social platforms don’t have.

Also, there will be other levers SN can pull to reward long-time users and consistent contributors over time (ex. referral system improvements, sub creator rewards).

The sats are only just beginning to flow.

⚡️

Curious–did you have some sort of data (user interviews, surveys, etc.) that you used to confirm that issue? Namely, "The big issue with it was that people don't attribute the value they put in with the value they get out."

Mostly curious about how you determined that, and how you think you'll measure success: "This change tightens that link."

Yes, pretty much no one I talked to before interviewing users understood why they were getting rewards. Also in @kr's recent interviews, none of them understood it either.

Most thought everyone was getting them, didn't even notice them, or thought it was some kind of magic determining they were worthy.

Measurement will be tricky. I'll mostly be watching behavior/looking at the stats. Are people engaging in incentivized behaviors more?

We'll probably eventually want to tell people precisely why they got the rewards they got. e.g. "You were 1st to upvote this popular Item. You posted this popular item. You contributed this popular comment. etc."

That's an interesting point, but the absence of reward isn't really punishment.

Mostly because if I trust the content you share (by upvoting it), it doesn't mean I trust your upvotes.

This makes sense.

In this release, trust is gained when two users upvote the same piece of content and diminished when one user upvotes something the other user doesn't

Does this mean that participating MORE in Stacker News can make the trust LESS? So it isn't only additive anymore?

Honestly, It isn't important to me ... the thing should work even if it was a complete blackbox. It piqued my interest from a programmers perspective.

107 sats \ 7 replies \ @k00b OP 7 Jul

If you participate poorly, yes you will be trusted less. Also if you don’t participate at all, you will be trusted less. It’s only by participating well that you’ll be trusted more.

0 sats \ 6 replies \ @dad 30 Jul

Will we be notified if we are deemed to have “participated poorly”? Or will it be a back room deal? Also, we need a delete button for our own content.

Back room deal is kind of derisive and inaccurate. The deal is made in public by other users. There’s no punishment. You just aren’t rewarded.

I’ll be creating a thread where everyone can request enhancements on Monday. I’ll add deleting (likely limited to within the edit window) to the list.

100 sats \ 3 replies \ @dad 30 Jul

FWIW, there should be a consideration for just sheer contribution over time because some of us don't have that 'x' factor that gets upvoted

This is good feedback. I’ll be revisiting rewards regularly. Is there any type of specific activities you think are valuable that we’re missing?

500 sats \ 1 replies \ @dad 30 Jul

Literally just volume of activity. I really like to comment on posts that don't have much commenting to feed the community some attention, but it doesn't get any kuddos or clout. I live on the /recent page when I'm on SN just to catch the little things

That’s very helpful. I’ll try to use your profile when I model rewards next and see if I can get them to overlap.

100 sats \ 0 replies \ @dad 30 Jul

Gotcha, so "trusted less" doesn't mean "downvoted" or anything. Thanks man. I've been contributing a lot (or feels like it) and haven't made more than 100 sats in earn for a few weeks so it just feels like grasping at straws. I know we're taking about fractions of cents, but it's more the mental "appreciation" token than anything

240 sats \ 1 replies \ @Jel 8 Jul

I think encouraging upvoting is a good thing. It helps to moderate the content on SN and supports the ecosystem behind SN.

I do not know if it is a good thing to make a split between posts and comments. Is not posting too much encouraged now because 1/3 is reserved for that? Obviously, there will be more comments than posts, so the rewards on posts will be much higher. Isn't there a risk that their will be too many nonsensical posts for the same topics?

It is great that this model is through so well. There is much more intelligence behind SN than the UI suggests.

Excellent analysis. I think you're right that comments will be more competitive and it might be better to make posts more competitive.

Overall it's an experiment and we can tweak all of this stuff overtime.

135 sats \ 0 replies \ @FatBear 8 Jul

This announcement has shown me how much I seriously underestimated the complexity of SN! On the surface it looks like a pretty simple message board, but of course there's algos running behind everything!

Great work.

100 sats \ 1 replies \ @moon 8 Jul

you only gain upvote trust if and only if you upvote well according to other users

Do you guys consider then the order in which the upvotes were given? Because if not I can just upvote posts that already have high upvote count and that would increase my trust -- by having herd behaviour. Whereas if my trust only go up considering the upvotes that come after, then ok, they are indeed validating my upvote -- then the system would be rewarding good curators.

From above, regarding upvotes of either posts or comments:

Users are rewarded based on:

  • how early they upvoted the content relative other users
  • how much they tipped
100 sats \ 1 replies \ @Majjin 8 Jul

Polls sound really interesting! Will it cost anything to vote?

Yes, user who creates the poll will likely set how much. Sats will, initially, go toward rewards. We might eventually do some staking instead, but I find the UX of staking pretty bad.

100 sats \ 0 replies \ @invar 8 Jul

went fromm 400 sats to 70 in the earn thing :)

26 sats \ 3 replies \ @jeff 7 Jul

36 hours feels like a low time preference experiment.

It could turn new into a roulette wheel like place to hang out. No real downside to getting it wrong, if in 36h it's all reset.

The downside is the posting/comment cost (admittedly low atm), and if you upvote poorly, your trust reduces as does your influence and upvote rewards.

I tend to upvote comments and content that isn't widely noticed or understood. I don't think this behavior should be discouraged with demerits otherwise it encourages cliqueiness and dog piling and discouragages discussions between individuals that are "paying" each other for their helpfulness.

That behavior isn't discouraged or punished. You and your discussion counter-party can send each other sats and it has no effect on your trust. If the discussion is in fact niche and no one else appreciates it, you won't get paid rewards, but that's about it.

200 sats \ 6 replies \ @gms 7 Jul

Interesting, it discourages commenting because yes and without contributing anything.

For my part, I go in every day to see if there is something interesting to read and I win some sats. It will be time to see what changes with this new algorithm. 👀

Why does it discourage commenting?

0 sats \ 1 replies \ @gms 7 Jul

By common sense I would say, if you comment and do not contribute, you would not win anything since I do not think they will give you an upvote.

Would this affect the trust factor or am I wrong?

Yes if you don't comment, you don't get rewards.

Only your upvoting affects your upvote trust.

You mean writing posts about more niche and more out there content harms my own trust? That would apply to me as well, I made quite a few more offtopic and out there posts.

Comments deeper in a thread discussing something as well.

If you comment or post and no one upvotes it, it doesn't harm your trust ... it only means you won't receive reward for it. You don't get punished for it - simply not rewarded.

thanks for explaining this, I was worried that niche topics could be harmful for your trust

I would love to have a two separate sections one for Stacker Talk/Discussion and the other for Stacker News!

ʙᴜᴛ, ɪ ᴛʜᴏᴜɢʜᴛ ᴛʜᴇ ʀᴇᴀʟ ᴅᴀɪʟʏ ʀᴇᴡᴀʀᴅꜱ ᴡᴇʀᴇ ᴛʜᴇ ꜰʀɪᴇɴᴅꜱ ᴛʜᴀᴛ ᴡᴇ ᴍᴀᴅᴇ ᴀʟᴏɴɢ ᴛʜᴇ ᴡᴀʏ...

⚡⚡

Idk... That trust model sounds like it's going to centralize participation really quickly. Is that what the 36 hour interval is for?

What's the upward mobility look like?

The 36 hour interval is for rewards only. You also don't have to be trusted to receive rewards.

Trust is a web and trust spreads pretty well in the web. It's true that if you aren't participating I don't trust you - but if you aren't participating, how am I supposed to determine you are trustworthy?

100 sats \ 1 replies \ @v1ntage 8 Jul

It's just the adversarial thinker in me, but do you think that mod-centric trust will lead to you getting a majority of the website's attention? And by extension, the posts and comments you give attention? In a non-direct, way, people get paid to interact with those they can infer are trustworthy. And you own the algorithm for determining that.

(I'm not criticizing. I know you're more interested in the longevity of the project. But it is a point a buttcoiner could make.)

I think there will bias toward the "mod's" POV, yes, but it's a rather soft touch relative to outright moderation. Currently there are also many users who have a trust level similar to my own.

We do plan to have each user have their own web of trust eventually - meaning they are the mod for their own SN experience.

The alternative to trust webs are trustless ranking algos, which we also plan to experiment with, but those are hard to design in a way that lets users posting content earn.

0 sats \ 4 replies \ @pony 8 Jul

so they are going to be a few get everything

It’s an experiment. If it doesn’t make content better, we’ll change it

28 sats \ 2 replies \ @pony 8 Jul

it encourages upvoting things only because a lot of people did/a lot of sats were given. so if there is yet another ama all users give these people their sats only to get a fraction back

lol

Oh, that's right, I didn't think of this.

Upvoting a post/comment that has already been upvoted guarantees that there are at least 2 users upvoting it. That could discourage being the first person to upvote.

You get rewarded significantly less if you upvote an item later. Early upvoters are highly rewarded. Your upvote rewards are basically proportional to 1/x where X is 1 if you were first to upvote, 2 if you were second, and so on.

how early they upvoted the content relative other users

I like the idea. I am more concerned about low effort altcoin posts tbh.

Or "TO THE MOON!" types of posts.

This is becoming a greater concern as time goes on. Best way to fight it is to increase the posting cost though. I’m not sure what the appetite is for that