You are viewing a single comment's thread from:

RE: Introducing UserAuthority (UA), @steem-ua and UA-API !

in #ua6 years ago

Well, this looks extremely familiar. It's the same curve which describes the distribution of SP across the population, except that it doesn't flatten out at the end quite as quickly.

Taken as a sum, how far down the curve in number of accounts before half of the population has half of the total UA ranking in play? About 20%?

Sort:  

This kind of curve is the product of nature, you can't get a normal distribution in anything when human action is involved. Pareto distribution is all we'll ever gonna get.

A Pareto distribution is natural; it is normal. You see it outside of human activity as well; anytime that there is an advantage to be had in an agent population by having more of a thing to get some of the thing, you get Pareto longtail distributions.

A lot of the discussion around cryptocurrency and especially around steem seems to go in a direction which denies that Pareto optimizations are desirable or necessary, and flying in the face of logic and reason like that is a good, solid way to fail.

So – we have another Pareto-reflecting curve on our hands with an exponential top-end and an exponential bottom end. What we really need is some way to visualize the difference between the rankings of the accounts involved in the traditional Reputational curve and the ranking derived from UA, and whether or not they are significantly different in ordinal rank for any given account or whether they, by and large, are in as much lockstep as they appear to be.

Because if it's just Rep but with a few accounts left out, it's not really very useful or revelatory. If it's a different Pareto distribution of the account list, how much different is it?

We could start to figure out how different they are with some pretty simple tools. You could start with a simple diff of the ordered list of all of the accounts by Rep next to all of the accounts ordered by UA. If there are transitions or transformations, that would be an easy way to start getting a handle on how to visualize them. (I'm thinking in particular of diff algorithms which let you know for any given member/line where/how much it has been moved, which is really all we care about. Magnitude of move and direction of move should be pretty easy to depict in some sort of graphical way.

After all, we know that UA takes significant computational resources to calculate. It has to be updated, effectively, for accounts of interest, every time there is a transfer within the event horizon of any given account. That means there's a significant cost for calculating UA for any given account. If it doesn't really do better than Rep for the vast majority of the account list, that's a lot of wasted energy.

So this is the kind of thing we need to see in order to determine if this is actually a useful metric or just another way to write Rep that takes longer and kills more electrons.

One example where UA shines is in - for example - recognizing @thejohalfiles as being influential. Without blogging an account cannot get upvoted nor increase its Rep score.

It's interesting that you would use that example, because it supports my position that follows are definitely not useful as an indicator of quality.

This account has no blog posts. It's influence is not related to its interaction with the social network in an observable way. While it does have a number of comments, none of them are particularly revelatory or insightful beyond a base level I expect of all people who interact with others on the social network.

So there's no reason to think that this one should "shine" at all.

And that's a problem, because we as users – no matter what we want to use UA for or what we imagine it might be used for – need to understand the reason that one account may be higher rank in UA than another. As it stands, and as the explanation and descriptions have changed, UA becomes an ever more amorphous single number attached to accounts which, in some way, at great computational expense, provides a number. One which doesn't come with an understanding of why that number is what it is, one that hinges on an interpretation of the platform mechanics which is unintuitive, and one which appears to be fairly readily gameable by engaging in behavior which isn't in the best interest of the way people are using the social network.

I think those issues are a problem.

But first we have to have an understanding of how UA differs from Rep in a real sense, in the context of comparing the two spaces as they stand – and you clearly have all the data necessary to do exactly that at this point, so let's do that.

After that, we can talk about what is going to be necessary for UA to be a meaningful designator, in part by allowing the system to give feedback to a user about why the ranking is exactly what it is.

We know exactly what Rep is all about and what it hinges on. It's about getting voted on. Stuff you make gets votes, your Rep goes up, it's a very simple signifier (even if it has some very obvious flaws as a comparator).

UA is a black box, and the things that you've said about what go into making the black box tick don't really jibe with creating a useful singular ranking of accounts for the purposes of a user looking to discover content, which is the one thing that it should do.

One example where UA shines is in - for example - recognizing @thejohalfiles as being influential.

Ultimately, on the issue of UA, we have to ask what "being influential" means if it's not about actually blogging or not about actually engaging people in comments. What kind of influence are we talking about? Because if it's just "this is an account that throws around a lot of SP," we already have very clear rankings for those.

So there's no reason to think that this one should "shine" at all.

A good reputation (UA score), being highly regarded and thus having many followers, can also be based for example on being a good developer, a wise witness or a precious curator.

And finally, if there is no logical reason at all for having a high UA score, then many people could simply decide to unfollow a certain account so that its UA would decrease.

Hold up. Let's be frank. There's no way that a "precious curator" is going to have a high UA score, given the things that are defined as the inputs into the algorithm which drive the ranking.

A curator has no reason to be followed, and so they don't tend to be. They don't tend to delegate out large amounts of SP, they don't tend to leave a lot of comments (because curation is hard enough as is, and unless they're a bot or bot-assisted they don't have time thanks to the fact that Steemit makes content discovery and uphill battle), or in any other way interact with the blockchain in ways which this metric considers useful.

A fact which has been pointed out to you by other people as well.

No one will stop following someone that they have on their list because they think their UA score is too high. That is ridiculously stupid and you should be embarrassed for even saying that in public. People stop following an account because they write or publish things that the follower doesn't want to see, so they stop following them so they no longer see that content.

A lot of this is just coming down to you guys saying, "UA is an important metric because we say it's an important metric and it measures lots of things and you're going to want to change your behavior because it measures things."

No. That's not how this works.

How is UA better than a near trivially computed metric based on an accounts number of followers divided by the number of people that it follows? How does give me any more information than just that? As a user, I know how the simpler metric fails, and I know how it succeeds and what it does. I understand what it communicates. UA does none of that.

I like a good fever dream as much as the next guy, but this is silly.

I see no reason why a curator shouldn't have a high UA score. Especially if UA could have an impact on voting weight in future. In my recent article I mentioned the example formula:

vote_worth = UA(voter) / UA(average) • SP • vote_strength.

If a high UA score strengthened the voting value of a curator (who may upvote your articles), then of course it could be worth to follow him.

No one will stop following someone that they have on their list because they think their UA score is too high.

I will. And I did already. I unfollowed inactive accounts as well as accounts which make much money with less effort in my opinion.

Somewhat related: thank you very much for defending me with your full-power vote in another discussion thread, I haven't got such a strong upvote for a year, literally

PS: This is a log-scaled UA Score distribution, on a 0.00 - 10.00 scale, where it gets increasingly harder to go from 7.00 to 8.00 than it is from 2.00 to 3.00.

PS2: this is merely a UA representation right now. Things could change drastically now that UA is out in "the wild", making people more aware of who to (un)follow.

Right, which is pretty much in line with SP in that sense. It might be worth plotting this with Y scaled log to flatten it out more for differentiation.

In fact, it'd be an interesting comparison to see how UA and SP scale accounts in different orders, or by how much it does. At some level, it's already described as a bit of a proxy. The question is whether it is ranking things significantly differently or largely the same in aggregate.

Also consider , when relying heavily on follows it will inevitably lead to a black market of -

Will follow for X $ for Y Days $

If anything, paying for follows will be considerably easier than paying for votes. Follows cost nothing. And while each additional follow scales the overall value of your following by the number you follow, despite all the effort on the front end to screen for isolated island networks and the like, it will be extremely easy to build inheritance architectures of follows that fairly straightforwardly game the system as long as one or a few witnesses jump on board, and given that there are witnesses who are heavily involved in the bot community as is – why wouldn't they?

I really appreciate the effort by devs but the criteria needs to be really diverse for the rankings to be meaningful in longer run.

From my previous comment, factors such as :

average number of votes received for each post , how soon someone replies to their comment , uniqueness and length of comments made/received , unique average posts made each day ?

Hello, I do not think I have run across your account before. So we may run in entirely different areas of steemit. A comparison of our UA may help me understand what the number means.
According to the UA web site, my stats: "score is 4.009 and you have a ua rank of 3389"

 MeYou
Rep5859
Followers1408539
Following13368
SP720680
AgeAug-17Nov-17

Fairly close numbers. Most of my content is just new user related, recently a few photo post, and for most of Jan-July a steemit game/challenge. So basically average to below average post quality.

Or you could spend five seconds looking at the stuff in my feed, and actually learn something useful and interesting about what I write and what I do. Something far more useful than a single numeric metric which provides no context, no real information, and nothing that actually reflects your interests or your personal inclinations.

I could tell you what my UA is, but all that would be is one more number floating around in the air. Like a stale fart.

And that, in particular, is one of the real failure modes of UA. It doesn't provide us, as users and as consumers, a tool for finding things that we want and will enjoy. It provides no meaning at all. If you said, "I like to write about video games, role-playing games, and game theory," I might say, "me too! I'd like to read what you write!"

But that's not a number. That's communication. That's content. That's useful.

And there is the problem. This provides us nothing. Nothing at all.

As tools go, this is purposeless.

It's illustrative that nothing involved with UA has anything to do with content. Nothing at all. And it's content that really matters to us as users and consumers, and even as creators. What we create, what we enjoy, and what we want to see our primary and it doesn't matter how many followers someone else as, how much delegation they do, or any of that garbage bullshit crap.

All that matters is content.

These numbers don't matter at all.

Ultimately, that's the problem.

After reading through all the comments, I still do not understand the purpose of it, or the use of followers, or the use of following, as being part of the metric. I have a list of 30+ Authors that I do not follow, but when I want to find a "story" to read I can go down my list and see what they have that I might be interested in reading. I do not read stories everyday, and have other interest also. But with this new system, if it becomes a tool of vote discrimination or content discrimination, am I going to need to increase my number of following to include the top 100 on the UA list, we all put a lot of garbage out and I do not need more in my feed than what I already have.

A close look at the top 100, and I am not sure I want a high number.

Technically, you won't want to consider increasing the number of things that you follow to increase the top 100 on that list, because they would be receiving a small portion of whatever it is that you generate that's interesting to the system that provides a mechanical advantage, and unless they in turn decided to follow you, the directed acyclic graph of interactions is definitely going to favor them, as part of that energy passes on to those that you follow. You get your power from those who follow you. Ideally, you want more people following you than you follow in order to have enough energy to make a difference.

What you need to do is to encourage those people who are in the top 100 of the you a list to follow you, because that would maximize the UA that you collected for a higher rank.

The cynical mind suggests that UA is really trying to determine a fairly complex thing about a fairly complex set of metrics:

"Whose ass do I need to kiss around here?"

Consider the list of the top 100 holders of UA at the moment. They are known witnesses, people with vast amounts of SP, people who are known to be willing to extend that SP in the pursuit of an ongoing project, and who have a habit of delegating SP on and off.

It's an ass kissing sensor.

It uses a new metric to define a new set of whales (look at the new whales, same as the old whales) to whom you must appease in order to get ahead in things. And that's all it does.

We have one more, slightly different, set of people whom it would be wise to kiss the ass of. Oh boy! Just what I always wanted!

But that's not a number. That's communication

Those are 2 different problems , recommendation engine vs curation engine.

Like a stale fart.

It's aged artisanally , not stale...

Those aren't even remotely two different problems, they represent the two things being done to try and solve the singular problem of recommendation and curation, which are the same thing. Pointedly, UA does not purport to be a "recommendation engine" or "curation engine" and really only supports its own voting engine which is cranked by effectively its own SP and issues out up votes to people who have registered with it and offered it enough material.

That will earn you some votes with the percentage that a a complex algorithm (you weren't expecting that were you?) decides you can have based on the amount of SP that you have given over into their care. You know, just buying votes in a slightly different way.

A recommendation engine or curation engine would care about the things that I am, personally, interested in and want to look at all the data that's coming in, see if it fits, and if so give it to me. That's not what we're getting.

If nothing else, this new UA algorithm appears to focus entirely on trying to drag systems which were never meant to do what they're doing into doing – this.

Sorry , I don't follow.

Recommendation is based on interests and Curation is generic authority based reward allocation.

Yes, you can say recommendation can take an input from curation , that is , recommend something with high authority and matching interest.

To me , they seem different yet connected problems.

Hey @shaka ,

tks for the upvotes...fam