RE: Finding Communities: Social Network Analysis on Steemit
I think your notion of allocating rewards to individual voters is inherently invalid. Rewards are awarded by the collaborative process of all voters (including both up and downvotes) working together. This even includes voters on other posts that occur at the same time. Interesting research though. Upvoted.
EDIT: Also questionable is stability of the results over time. My two accounts show two different authors yet the accounts have voted in an identical manner (automated to behave as a single account) since July if not earlier. Only in the initial period was there any difference (second account was generally not voting at all). I have no idea to what extent this carries over to the rest of the data but given massive changes in usage since the start, and concentration of rewards during the peak market price period of July-August, I think significant.
That's true, so I used voting counts to cluster communities. Instead of payout(vshares), rshares is more accurate measurement, but payout still have practical implication because it shows usage and allocation of resources (rewards)
In practice what it shows is mostly what happened during July and August. With 95% price (reward pool) decline (and similarly much lower price earlier), the weighting is extreme, and largely excludes most platform usage. Good that clustering is done using rshares.
IMO, clustering by rshares less makes sense because the variance is very high and as a result all whales can be into the same group. (that is, it make big stakeholders closer). But when I have a change, I can do it. Thanks for your feedback!