Building Better Bots: Ranking the Top Curators
Food For Bots
When it comes to curation rewards, Whales (and Dolphins) are in fierce competition. Ultimately, this boils down to an information game, making the smartest choices you can with your votes, based upon the rules of the game. @dantheman recently posted his thoughts about this issue, where he mentioned that this competition requires Whales to use Bot Algorithms. Further, he described the role of Dolphins and Minnows as being one of generating data for bots to use.
When bots are attempting to identify content to up vote they will take clues from minnows. Good dolphins can even start to earn curation rewards when whale-bots pick up on their usefulness as an early indicator. - Dan
In this way, Minnows and Dolphins who vote accurately, can actually begin to have a large impact indirectly. Some whales will begin to tune their bots to vote on everything that top curating Minnows and Dolphins vote on. In this way, there emerges a greater incentive for Dolphins to vote accurately, since having posts upvoted by Whale's Bots after you upvote can yield modest curation rewards.
Creative Commons: Source
Note: This is probably mainly a game for Dolphins and note necessarily minnows. Because of the relationship between curation rewards and Steem Power, an account must have at least a certain level of Steem Power to achieve curation rewards, regardless of how much skill you have when voting.
Quantifying Curation Skill/Ability
The question then becomes: how can we quantify Dolphin's and Minnow's curation "ability"? Here I present a simple technique for quickly estimating someone's skill at identifying valuable content. My hope is that this information will prove valuable to bot designers. After all, as Dan said, an "evil-whale" is one who makes easy profits due to lack of competition. Since this is ultimately an information game, I want to present these results to the whole community, with the hope that publishing up to date, relevant information will increase competition, and decrease the odds of emergent "evil-whale" behaviors.
It is not appropriate to simply compare an accounts total curation rewards over a fixed period of time. This is because curation rewards scale with N Log(N), where N is an account's Steem Power. A single random vote from a whale could easily yield more curation rewards than 100 perfectly-timed, intelligently-placed minnow votes. For this reason, we must normalize each user's curation rewards (over a certain period) by the value of N Log(N) for that account. This normalized value then gives a simple numerical "score" to describe users' curation abilities.
Show Me The Data!
Since I like chunking through data, I have determined a curation "score" for each of the top 500 accounts on steemwhales.com. The reason I limited it to this range (Whales and Dolphins), is partly because acquiring the data was taking a decent amount of time, but mainly because I assumed that only these larger accounts would have a sufficient interest to put forth the effort to play the game and try to maximize curation rewards. Still, I do not discount the possibility of smaller accounts having greater curation "scores". If this type of data proves valuable to the community, I will attempt a more thorough characterization of all accounts. Still, it is likely that the top curators I have identified would still be top curators if more minnows were included.
The following figure shows the data, ordered by curation "score". The first column shows the user's whale rank according to steemwhales.com, the second column gives the username, the third column gives the account's Steem Power as of a couple hours ago, the fourth column gives the account's total curation rewards for the past week, and the final column gives the normalized curation score. This was calculated by taking the curation rewards divided by N Log(N), where N is the account's Steem Power, and finally normalizing by the maximum score, to give a max score of 1.
What Can We Learn From This?
- NO super whales are super curators. Out of the top 30 curators, the highest-ranked whale to appear is @wang, who ranks 27th on the Whale scale.
- The majority of high-performing curators are modestly-sized dolphins. The median Steem Power of the top 30 curators is about 27,000, which is probably not in Whale-territory quite yet. These slightly larger dolphins have an incentive to play the game and try to gain curation rewards, and we see that in the rankings.
- @recursive2 is ahead of the competition. This account is probably owned by @recursive, who is in whale territory with 192,000 Steem Power and possibly has a clever Bot working for them to capitalize on curation rewards. Well done!
- Keep in mind that a week is still a fairly short time period. I have seen my weekly curation rewards range by up to 100% or more as I try to find valuable content. These rankings are probably good estimates of curation ability, but they are obviously subject to fluctuations and random chance.
Bots Need Humans
One very important takeaway from these results, is that humans can outperform bots. I was very happy when I discovered that I am the 6th highest-ranked curator (based upon this simple metric). However, I was even more surprised to see that I had beaten out @wang, by more than 60%. This means that if @wang switched to a far simpler voting algorithm that simply immediately votes on everything that I vote on, he could increase his curation rewards by almost 60%! This is what it means to provide valuable information. Dolphins and Minnows can provide this type of information to Bots and their operators, providing a valuable service, for which they can be rewarded (by having the content that they upvote be immediately upvoted by Whale's bots).
I keep meaning to devote a day to doing nothing but curation, as an experiment, but life keeps getting in the way.
I think your post might also stand as a bit of an experiment as to how much you can earn through the efforts of minnows and dolphins alone!
Thanks, and lol yeah, I'm up to what, like ~670 upvotes, zero flags, and $18, haha. Oh well, everything's an experiment at this stage :)
Something very interesting about that :)
For the record, I am NOT Sybil attacking this post with upvotes, lol
I don't know who is doing the voting, but they are helping you in that I wouldn't have seen your post in the hot list without them. I expect it is the same for others. So minnow bots do have a value. You are now on 893 votes, no flags and $35.
There is an account "steemitmarket" which has ~800 spam accounts and is trying to buy/sell votes. It seems like they decided to test out their whole bot army by spamming this post.
This post is superb. I haven't really thought about the bots using the humans' actions to create future trends. I'm already thinking a bit more forward. Thanks.
Thanks! Yes I think more sophisticated bots will begin to use human actions as indicators to affect their voting trends. Forward thinking indeed :)
First, thanks for taking the time to create this informative and valuable post. However, I see a very real problem in all of the post like yours. The info is presented to an audience of people who are very advanced in their understanding of what is going on "behind the scenes" on Steemit and with how the system works.
I believe that , from what I have read in the White Paper, this platform has the potential to quickly grow past even Facebook and its more than a billion users. This will definitely not happen if Stemmit continues to be an "programmers or hackers only" elitist oriented platform. People like yourself have a very real opportunity to profit beyond your wildest dreams by providing information in a manner that is useful to people outside of your inner circle.
For example, I have yet to see a post that clearly shows a clear path for a newcomer to Steemit who consistently posts quality content to garner attention by virtue of the quality of their posts.
Why? What are the guidelines and or parameters for quantifying quality on Steemit and how does that quantification system work?
For example, if the majority of current participants on Steemit love the art of Picasso but throw up at the sight of anything from Van Gogh, then that would mean that Van Gogh as an artist would be deemed to be of no value as an artist on the Steemit platform.
In my humble opinion, the incredibly explosive potential for Steemit and everything "Steem" requires a mandatory ability for all "art lovers " to have the opportunity to experience the artist/style/content that they find pleasing and inspiring thereby opening up the platform as a source of the widest range of content from the "lowest" to the highest" quality as determined by the people interacting with the content. I use "art" here only as a means of establishing context. This would apply to any kind of content....or category of information.
I understand that funding a massive truly open repository with the kind of technical resources required to execute it would be a major challenge. Steem. in my opinion, has enough critical mass already to mount an aggressive funding initiative to realize the lofty concepts and goals outlined in the White Paper.
Thanks for your comment. You've got a lot of good questions, and the truth is, some of them might not have answers. For example, "What are the guidelines... for quantifying quality on Steemit...?" That's a really good question, but you could apply it more broadly to the internet in general. It's often the case that popularity and 'value' are found in unexpected (possibly undeserved) places on the internet. It's incredibly hard to predict what will or will not go viral on any platform. Remember, Van Gogh wasn't even discovered until after his death. As far as Steem serving as a platform for intellectual property like art, I know that there are still new ideas in the works as far as ways to reward people for storing their intellectual property long term on Steem. It will be interesting to see how everything evolves. Thanks again for posting.
I am new to the platform, I did a lot of research before joining steemit, if we are to dream big, and looking to compete with Facebook, and the rest, bots up voting and bots wars need to stop, also new comers like me will need to be supported.
Interesting ideas and helps with concepts I've been mulling over myself. My initial thought coming to Steem from Bitcoin and seeing value in both got me wondering about the strategies to benefit for someone who sees value and potential in the Steem platform but lacks the skills, abilities or inclination to participate in content creation or curation. If that potential investor were to power up and passively hold SP, they may still benefit from growth of the overall system, but dilution would mean they would miss out relative to active participants. Curation bots could give the potential investor without the skill, ability or inclination to curate the opportunity to "employ" curators to help them increase the yield on their investment.
The current reward system is heavily weighted towards material whose value is instantly recognised. I would like to see a system that also recognises and rewards posts and information that is useful over time, but not necessarily viral (e.g. reference material and detailed how to guides). I would be interested to hear of the ideas being put forward for "ways to reward people for storing their intellectual property long term on Steem".
I didn't expected to see whales on the list, because if you have more steem power, it is much harder to get in the first 1%. Also I don't think it is possible to make 1% return with curation.
EDIT I have just noticed it is per week and not per day. 1.5% / week is possible, I also made that target for myself. :)
recusive2 uses voting proxy. I don't really know what it is, but I think he might use other's steem power to vote (recusive).
Anybody knows what voting proxy is? :)
Voting proxy may be a new idea that's been thrown around which allows you to delegate other accounts to vote with your voting power. I'm not sure if it was ever implemented though.
This is something ive been wanting to see for a while. Ive flung myself into curation recently trying to figure it out, thanks for providing this data.
No problem! I'm trying to find my groove too. Thanks :)
Have you adjusted for creation rewards and powering up/down?
This just takes into account the total curation rewards for the last week and the current level of Steem Power. Powering down would not make a big difference, since it's like a 1% change per week. If someone has powered up this past week it would make their score appear lower than it should be.
It still doesn't benefit the minnow though. Huzzah, I can be a useful tool in identifying good content that other people can then monetize.
There is always the possibility that exceptionally gifted minnow curators can be hired directly by whales to pick out valuable content for them. In fact, this is already the case. I know of one whale who has hired at least 10 different users to pick out content for them.
Would that be @smooth ? Great post @trogdor Quite useful information on voting. I having wondering why more "Whales" don't up vote, especially when it helps them and the whole STEEMIT system.
@streetstyle
Thanks! and yes, I think @smooth has a waiting list of people to find good content for him. Not sure about any others.
Interesting analysis, thanks!
I agree that following smaller and more dedicated curators is a great idea of simple yet effective automated strategy. In its current basic form, algorithmic curation is a race to the bottom as bots tend to compete around the same authors that stand out statistically, and try to undercut each other by voting earlier and earlier on new posts from the same handful of "bluechip" authors. Right now it's still somewhat profitable as there is still a lot of herding around the same handful of authors. But increasingly, competition between bots results in pyrrhic victories due to the diminishing returns as the bot fight is getting closer and closer to the posting time. The amount of "alpha" available is also shrinking as people have started to realize that jumping late on the same bandwagon as everyone else isn't a good strategy and it's more profitable, constructive, socially useful and intellectually satisfying to find and promote unloved good content. As a result payout distribution is getting flatter and flatter.
As you have identified it, tracking smaller human curators is the next logical step as only human curators can really predict how other human curators will react to a specific piece of content. That's exactly what @smooth is doing, and I expect this to become a new trend. Rather than trying to extrapolate author performance based on historical cues only, I expect that bots will evolve to detect which curators are human and which are bots (based on amount of entropy in the voting pattern), analyse human curators performance, and use them as heuristics for their own voting. I quite expect automated curation to lead us in interesting territory with bots based on natural language analysis and machine learning techniques. For human curators, there are careers to be made. Whales will start following good curators which will improve their returns. And quickly they'll "scout" best curators so that they can benefit exclusively from their insight.
The bottom line is, as OP said, if you feel that you have a knack for finding good content, keep doing it regardless of how effective the returns. By doing so, you are building a track record, and positioning yourself on the soon-to-emerge market of curation-as-a-service.
Thanks for the thoughtful response :) I've also noticed the race to the bottom as votes get pushed earlier and earlier. It would be interesting to see the 'scouting' you mention start to become more frequent. I hadn't thought about quantifying the entropy in voting patterns to differentiate human and bot curators. That would actually be a challenging problem, especially as bots begin to imitate their scouted human curators more and more. Even with seemingly random voting patterns you would have to look at correlations with other accounts' voting histories. Really interesting!
Sure steemit does need some good robots because it is been a while since copy cats are copying and pasting same old stuff over here by making a few changes and robbing such a good platform like steemit. It should be stopped of course I like your theory about better robots thanks for sharing.
Thanks for the comment :) I think these tops of problems will get smaller and smaller as bots get smarter and smarter and imitate human actors more and more.
Nice information, thanks @trogdor
Thanks!
Very interesting article.
💋@halo 💋
Thanks!
@trogdor Not sure how this wasn't in my feed. But I've been working on a similar post for most of the day now. Guess I was late to the game on it. Glad I caught you randomly like that before I put any more effort in. Awesome work!
Thanks! Although this has kind of unravelled in a way I would have preferred it not to.