Hi @personz
- interesting read indeed. But it's very computational-heavy and one shouldn't use a cannon to shoot a fly (ref. Confucius). UA on the other hand is relatively simple and extendable.
- UA is not meant as a stand-alone metric. It's derived itself from the follower graph on the blockchain,
- UA can be used Steem-wide, via a reversed binary index holding all UA-scores, calculated by witnesses daily / per hour, where the UA-results binary's hash is stored in each blockchain block. Remember all the follower-data is in the blockchain already, just currently not in a usable way to read every user's UA on the fly.
- please note that with every individual follow-change, the entire UA matrix is changed. If user A decides to follow user B, that one follow changes the UA for every user. However, the impact of that one additional follow, is so small on the entire follower matrix, the difference for each account is roughly 0.0000000001% and therefore marginal. A daily / hourly re-calculation of UA is therefore to be considered OK.
- indeed, just because something seems difficult doesn't mean we shouldn't try: in fact the difficulties make it an even more interesting problem to solve.
- I haven't programmed in NodeJS for over 13 months, but I found a NodeJS dev willing to write an MVP, based on my two UA-articles, where I coach the dev. I'll look into NodeJS myself as well (to blend UA code into existing APIs).
- 3 essential steps are needed to let UA succeed:
a) develop the UA MVP (via Utopian), test-drive temp results, alter the damping factor in case needed. Since UA is a game-changer (who you follow and who follows you becomes important), users will change their follows accordingly after deployment. "New rules lead to different behavior**. That anticipated behavior needs to be factored into the MVP test-results
b) evaluate and discuss MVP results Steem-wide, propose and vote for HFs
c) deploy Steem-wide.
Let's begin with -a- I suggest
Thanks for your detailed response. I should say right off the bat, because I didn't say this before and it's relevant, that I do not agree with the usage of a "honestly" metric to factor into payouts or vote weighting, as you suggested in your first post. I see it's usefulness in providing facts (i.e. a score) for a human led action (such as a witness could do) to divest a user of their SP (not including delegated SP of course) and prevent social media operations (voting, posting, etc.), while allowing the wallet to exist.
Your proposed algorithm is a good indicator of dishonest accounts but it would not be good to enforce this view on the populous. I don't believe the view is comprehensive enough for that. So I would very much oppose the two hard forks you suggest, as you suggest them. And I can tell you that in it's current form it will certainly not be adopted.
On the technical aspects, your calculations only hold up (if they do hold up, I didn't check) for follow information, not when including the much more voluminous vote information. I think votes are a better indicator of "the network", i.e. the topology, than follows.
The reason why it would work well for the Utopian bot is that Utopian is very heavily moderated and has high and restrictive standards on it's members / users. This is not the approach for the Steemit social network in general, so I assume you would prefer to go towards that. I don't think that will fly.
Looking forward to the MVP and updates on how the algorithm plays out, especially profiler / benchmark information.
Indeed, that was one of my thoughts. In real-time, at current technical levels, it would be improbable that it could be done.
If the payout window stays 7 days, then I guess even every a 5 days recalc could be reasonable. With more users coming to Steem the amount of needed calculations would go up very fast. It would be a quite impressive piece of work I guess.
And if you never try, you'll never know.
I am scipio
?
;-)