In previous iterations of this list we took into account...
- Looking at the top 100 posts in trending and examine authors (dead end)
- Examining every vote that has occurred on each post until we get some fixed percentage (we chose 12% at first, but if you expand to 20% you'll get a much bigger list of curators to examine more closely.)
- Ranking each curator according to how many posts they got in on before they reached the treshold.
- Determining the standard deviation of rank (not covered because it turned out to be a dead end).
- Adjusting rank to take into account social network influencer effects by looking at the total rshares prior to them jumping in.
- Adjusting the final list by getting the average of rshares prior to them jumping in.
And now because everyone knows that 7 is a lucky number, here is the 7th and final iteration of the process!
What we had not taken into account in all of these efforts is actual luck. As in answering the question of
"How often does each user actually get lucky compared to how often they vote?"
As it turns out this was not the easiest question to answer from a technical perspective.
According to the API docs, there used to be a list of votes on the account when you queried the account. Imagine our surprise when we discovered that feature is now gone?
The recommended way is to use the databaseapi.get_list_votes method. However this gives you a mixed bag that is remarkably hard to deal with. It's fine if you're looking at a specific posting. But for anything else you literally have to walk through not only every single vote the person ever made, but also every time someone voted for or against something the same post as them. You end up diving into each post and every comment. It's a tar pit and for our purposes it was just to slow.
We found it was easier to examine the results of account_history_api.get_account_history with a limit of 10000 and a start of -1. The returned result is the last 10000 things that account has done or has had done to it. Voting is one of those things, huzzah!
From there we simply filter by votes in the last 7 days and bob's your uncle!
The end result looks something like this...
In the final list we just clear out anyone below 1% since 1 out of 100 is our arbitrary threshold otherwise the list gets very, very long.
The end result is this list.
So who's the luckiest the luckiest curator today?
It's hard to say.
@kushed comes out on top at 4.3% overall however he only has 4 things he curated that made it trending. If you were just following one person and dittoing their votes, @kushed should be your choice since he's correct about 4% of the time.
@cheolwoo-kim has a somewhat lower luck factor of 3.7% yet they've got 11 things that made it to trending also his average rshares are an order of magnitude lower and really lower is better on that metric. So if you're wanting to talk to someone to find out what their secret sauce recipe is, you'd get the most mileage from chatting up @cheolwoo-kim.
Every person in the list (along with many others) have been added to our own dittobot and we updated it to take into account luck. Now it's a lucky bot ;)
We're going just leave the algorithm alone now to let it run for a week and and see what happens to our curation. However we will post this list daily and once the testing is complete we will post the source code for your own tinkering.
In the meantime starting with this post. Each time we post a Lucky 7 listing we are setting the beneficiaries to the Lucky 7! We give each of them a slice of our pie equivalent to their luck ranking. So upvoting this post is helping them as well.
Thanks for reading and remember as (almost) always, this post is 100% steem powered up!