Also view your full post history in tabulated form!
My new App, Steem-Index tells you two values: your V-index and P-index. If you're an academic, you might think these terms sound familiar. You'd be right!
Background
H-index is a measure of how productive a researcher is, based on their publication record. It is a function of both the number of papers they've published, and how many citations each paper received. To calculate it, you just need to sort their papers by the number of citations each paper has received, and find the maximum value for which the number of citations is greater than or equal to the paper number. Simple, right?
Well, now you can calculate that value for your Steemit posts! Instead of citations, we use either the rewards your posts have received (Steem Power; Post Value), or the popularity of your posts (Total Votes; Post Popularity). I call these two measures V-index and P-index, respectively.
Usage
Fill in the form on Steem-Index for either your quick calculation (based on top-level posts that haven't paid out yet, i.e., posts less than 30 days old), or your full calculation (your entire post history, including replies to other posts). While the full calculation is more extensive, and more representative of the impact you've had during your tenure as a Steemian, it can take considerably longer to generate. It also doesn't include how many votes each of your posts received, unfortunately, so you are limited to the V-index in this case.
Under The Hood
The website is a flask-app (source available at: github/geoffrey-steem). Unlike my earlier app, this one is less ad-hoc in implementation, querying the Steem blockchain directly using @xeroc's excellent python library. While the quick method uses the get_state() command to find each of a user's recent posts, the results are limited to posts which haven't paid out for their 30-day period, yet. The /full/ method uses get_transaction_history() to iterate over every transaction a user has made on the blockchain. This is quite slow, so I store the results to a "database" (read python dictionary written to file), which can only be updated for a particular user once a day (to speed-up repeated access; and limit wasted bandwidth).
The quick method uses websocket to communicate with steemit.com, but the /full/ method requires thousands of requests, so a local copy of cli_wallet has to be ran with 'account_history' enabled (this necessitated increasing the ram on my seed node which I host the website off of).
Limitations
Both methods have their ups and downs. Neither is perfect (yet).
Method Advantages:
- Quick:
- Speedier
- Doesn't need a local instance of cli_wallet or blockchain download
- Contains both price and vote data
- This allows for both the V-index and P-index to be calculated
- Posts link properly to their threads
- Full:
- Considers all comments by a user ever made on Steem
- Stores tabulated results for access outside of the website
- Reward measure is accurate to what the post actually paid out*
*: The reward for the quick method is estimated based off of author_rewards, a value for which I could not find the exact calculation. Empirically it seemed to vary between 1800-2000 author_rewards / Steem Power, so I used that range in my calculations. While get_state() reports rshares, which other posts suggest should correspond to rewards based on rshares^2 / total_daily_vshares = Vests, with total_daily_vshares = 10^24, this didn't seem to be the case. Alternatively, rewards could be estimated from 1/2 the SBD payout (which get_state() reports), but that involves calculating the 7-day median price on the date of the post, something I'm not readily equipped to do.
But don't be afraid to try both.
Future Efforts
The ultimate goal is to create a high-score list of users with the highest impact. This will allow you to at-a-glance figure out who is really worth following! I'll have to iterate over all user accounts (there are 50,000+ now) of them, and keep it updated as new blocks are pushed. This will take me considerable time to implement and run, so for now enjoy the app manually, and give feedback while I flesh out the concept further.
Also, I'd like the fuller page to incrementally fill the table. I'm told this could be done using websockets, so I'll have to research that further. For now, it just continues to show the page as loading while it tears through the blockchain.
I'd also like to get more stats to play with. Since I'm becoming better at using the RPC client for blockchain analysis, it shouldn't be too long before I provide measures based on number of followers, reputation, frequency of posting... all kinds of cool stats!
Finally, kind of simple, but I'd like to allow the viewer to sort by different columns. Right now I manually generate the html for the table, so that'll require a little bit of rework.
Some Users of Interest:
Perhaps you're wondering how the whales score? Here are some of the ones I've looked up based on them hitting the trending page and having high reputation. Feel free to post more interesting or surprising users in the comments.
V-index
What do you think? Would you rank the prolificness of each author the same way?
In case what these values mean wasn't clear. Gavvet has 92 posts which have received over 92 Steem Power at payout. That's quite impressive!
Closing Thoughts
As is readily apparent, I'm not a web developer. I just really like Steemit, and am doing everything I can to take advantage of the idea of an open blockchain. This allows me to experiment with developing apps, and grow my expertise. This specific app is a considerable improvement over my first app, which dumbly scraped steemit.com user pages to get account balances. It also has a much nicer and cleaner layout, though there is still considerable room for improvement. I hope to continue to evolve my skills, and am very open to hearing any feedback you may have.
Very interesting. So I have a full V-index (reward value): 25. Problem is, I don't know what that means because I don't know what these mean:
I see you tried to explain that in the
*, but I just got lost a bit more between get_state, total_daily_vshares, and author_rewards.Sorry if my question is stupid as I'm not an academic or familiar with H-index.
My other two values are
V-index (reward value): 11
P-index (popular votes): 20
But I'm not really sure what those mean either.
Not a problem, I apologize for my terseness!
Let me know if you need further clarification.
Have you thought about modifying the V-index so that you don't compare to post payout value in Steem Power directly but instead to that post payout divided by some factor? For example, you could divide by the average payout value of posts at the time? Perhaps the average would be an average within some moving time window. And it wouldn't have to be a mean, perhaps a median or some other statistic on the distribution of post payouts within the time window would be a better value for the factor. Anyway, then you could saying something like, "If the V-index of an author is X, it means they have X posts that each earned at least X times the 30-day average payout at the time they were posted." I think that would be better because right now you are already sort of arbitrarily choosing a factor. And the risk is that if you choose a factor that is too large then everyone's V-index would be restricted to a very narrow range (possibly all 0 even) and thus be useless, or alternatively, if the factor is too small, then the V-index basically reduces to the number of posts of the author, which would make it a redundant metric.
Very good points. I had considered a few alternatives, perhaps I should use (regularized) rshares, as I originally was considering, since those2 represent the relative rewards a post gets compared to all of the others for that 24 (I guess 12 now) hour period, if I understand it correctly.
On your second point, empirically the formula I use seems to give very reasonable values for two cases: 1) accounts like mine, which have a very small number of popular posts, and a medium number of not-popular posts. 2) accounts with a large number of popular posts, but also a large number of posts. As I continue to try other cases, they seem to form a reasonable spectrum, but I'm open to tweaking it if a better metric is suggested.
To your original point, I'll need to go forward with acquiring the measure for all users before I can regularize it. Alternatively, I'll need a good way of figuring out the total-payout at an arbitrary time-point. I haven't figured out how to get anything more complicated than the agreed-upon past 30-day SBD values.
The extra parenthesis there trips me up. Is it 11 Steem Power AND 11 Steem Power worth of SBD (so 22 total Steem Power)? Or is that an "or" meaning "at least 11 Steem Power worth of value, even if it was paid out in SBD?"
I think I get the P-index now, thank you.
These things get complicated fast, so I applaud you for grasping the full meaning. I only count my estimate of Steem Power. I didn't want to deal with the variable rate of SBD conversion.
The implication of this is everybody's V-index is stricter, it's an under-estimate of their true reward values, by at least 50%.
Top two by rep (according to http://steemwhales.com/?p=1&s=reputation)
@stellabelle V index 114 Rep 72.3
@tuck-fheman V index 93 Rep 71.2
I'm currently a 1. (I think my academic H index got up to 5 or so, but I left academia awhile ago so haven't really checked lately)
It would be interested to graph the correlation between V index and reputation score for a larger data set.
Hilariously we posted the same idea at the same time. Check my comment below.
Interesting, when I calculate it for everyone, I'll check out the r2 value with rep. I bet it's pretty high!
My gut feel would be that outliers would be those who get most of their upvotes and rep from commenting rather than original top level posts, but it seems like there will be high correlation.
Those are actually factored into the /full/ method, too. Though, similarly, I would anticipate whales who don't comment often to affect the measure, since the reputation calculation is weighted by the amount of SP an account holds (I think?).
Nicely done! Looking forward to using this to assess myself (and hopefully track improvements over time), and others that I'm interested in following too.
Thanks for work on this, the more data on Steem the better :)
looks like tuck-fheman actually beats Gavvet with a V-index of 93! Any other even more successful accounts out there?
This is pretty cool, great work!
thanks for the initiative. love what I'm seeing for my account @ace108
gives me a platform to learn more about my post.
[
voted @geoffrey for witness. :-)
謝謝 (google translate). Your support encourages me to continue working on cool apps!
seems like a good idea
Written for the Russian speaking community
[Steem-Index] Анонс аналитического сайта - Насколько популярны ваши посты.