The Typology of Curation: What Kinds of Curators We Have Now?

in #stats8 years ago (edited)

Two days ago, I posted How Much Do Whales Influences on Rewards?. After posting it and having discussion with others, I found this study can provide more deeper understanding of curation patterns. As a follow-up study, this post is focusing on typology of curations: grouping curators by certain patterns.

Method

I used k-means clustering to classify the top 100 curators (by influence) into four groups. The sample's total influence is 99.5%, which means almost all of curation is done by the top 100 curators.

Data and Variables

The data is from Steem blockchain from 2016/8/16 to 2016/9/10.

  • Influence: Percentage of payout a curator used
  • Power: Percentage of Steem Power a curator owns
  • IPR: Ratio between influence and power (Utilization of Steem Power)
  • N: Number of votes a curator casted during the study period
  • Range: Number of unique writers a curator voted for
  • Mean: Average number of votes per writer
  • Max: Maximum number of votes on a writer repeatedly
  • StDev: Standard deviation among a curator's votes on each writer

The following table presents some descriptive statistics about the four groups.

GroupSum_InfluenceAvr_InfluenceSum_PowerAvr_PowerAvr_IPRAvr_NAvr_RangeAvr_MeanAvr_MaxAvr_StDev
115.860.5130.330.980.52120.51123.710.975.970.90
246.991.2018.110.462.59582.55329.721.7724.092.65
331.831.3810.820.472.94802.70260.963.0861.936.58
45.270.751.350.193.901410.57164.578.5770.0012.08
Total99.95-60.61-1.65547.91238.482.5130.393.67

1. Lazy Lions

Curator in this group show low utilization of Steem Power on average (0.52), but they still have a great potential to become influential curators since they have more power than the other groups (0.98). The list of curators is as follow.

AccountInfluencePowerIPRNRangeMeanMaxStDev
@dantheman3.412.541.342223400.65101.24
@jamesc2.723.190.851501351.11141.86
@rainman1.801.491.212031711.19101.31
@ned1.565.720.27452540.1830.37
@val-a0.882.800.3226770.3320.49
@fuzzyvest0.570.580.973062141.43121.93
@dan0.444.490.101120.0910.36
@witness.svk0.420.540.781691301.30101.24
@wackou0.340.620.5592571.61101.59
@skywalker0.280.630.4418131.3830.62
@laonie0.270.141.87123012271.00171.32
@blackjack0.261.000.2615121.2520.43
@roadscape0.260.650.4054600.9140.77
@lafona-miner0.250.280.912011481.3681.00
@firstclass0.210.660.3117121.4240.86
@arhag0.190.610.3130400.7530.60
@freedom0.180.530.3417121.4240.86
@cryptoctopus0.170.151.131471111.3391.17
@proskynneo0.160.540.3059820.7220.44
@datasecuritynode0.160.440.3621181.1730.50
@books0.140.160.8517131.3140.82
@disney0.140.150.9216131.2320.42
@bmw0.140.180.7417131.3140.82
@samupaha0.140.200.67-2241-0.01172.24
@alphabet0.130.330.4015141.0720.26
@excalibur0.110.320.3615141.0720.26
@analisa0.110.400.27331.0010.00
@arsahk0.110.390.28431.3320.47
@bitcoin20160.110.420.2517121.4230.76
@delegate.lafona0.100.101.032011481.3681.00
@cryptogee0.100.081.254112361.74121.79

2. Hidden Gem Explorers

This group consists of the biggest part of the total influence (46.99%) and they have the highest range and the second lowest Max and StDev. This implies that their votes are dispersed to various authors.

AccountInfluencePowerIPRNRangeMeanMaxStDev
@blocktrades11.293.713.046304861.30221.51
@berniesanders10.412.973.509977871.27151.71
@tombstone3.621.871.943211961.64122.18
@summon3.001.292.323642211.65141.99
@complexring2.490.673.699585441.76403.08
@nextgencrypto1.750.513.4110617591.40141.51
@pharesim1.640.871.888374062.06543.93
@silversteem1.550.463.3713817601.82152.08
@riverhead1.550.692.255183361.54131.41
@steemit2001.540.562.744381782.46142.35
@satoshifund0.970.352.783291302.53213.61
@au1nethyb10.780.431.844172092.00111.81
@silver0.710.193.7614037631.84152.13
@badassmother0.690.262.686622372.79162.83
@donkeypong0.470.251.904812861.68152.02
@joseph0.450.192.373741552.41303.63
@xeroc0.350.181.934782401.99453.48
@rossco990.300.151.992101191.76131.56
@justin0.250.073.6010977741.42141.50
@clayop0.250.560.44209782.67274.53
@slowwalker0.240.083.198954372.05283.61
@cass0.210.102.044542192.07202.53
@boatymcboatface0.200.063.441851111.67111.37
@james2120.200.092.205702312.47294.21
@steempower0.190.121.664952152.30373.62
@onceuponatime0.190.280.683801981.92433.88
@sean-king0.160.131.223992281.75322.68
@dollarvigilante0.160.072.296222392.60464.23
@steemservices0.160.053.139217801.18142.66
@stellabelle0.150.210.744772431.96444.27
@renohq0.140.131.072701132.39152.51
@anwenbaumeister0.130.071.797714251.82282.40
@toxic0.130.052.401731071.62121.48
@kevinwong0.130.062.209734722.06423.08
@steemychicken10.120.150.812591491.74192.63
@the-alien0.110.042.542472481.0080.77
@steemrollin0.110.071.513322311.44141.81
@inboundinken0.100.032.965392572.10202.20
@liondani0.100.091.115942922.03604.44

3. Village Builders

While this group has a significant influence on payout, they votes for small groups of writers repeatedly, probably due to build their own communities (they have very high Max and StDev). A good example is @abit and @boombastic who support the Chinese community. They have relatively high possibility to be developed into their own DCGs suggested by @ned

AccountInfluencePowerIPRNRangeMeanMaxStDev
@smooth13.192.864.627913432.31657.50
@steemed4.071.602.54298803.72456.54
@itsascam3.941.532.57297793.76456.60
@smooth.witness2.290.534.307823272.39657.63
@steempty1.280.701.844781652.90344.83
@kushed1.170.502.346572512.62857.40
@xeldal1.150.651.785972042.92485.65
@enki0.910.531.725601743.22465.58
@steemroller0.460.182.57295734.04446.69
@abit0.450.640.717832013.90607.48
@cyber0.430.114.0220584504.57668.08
@boombastic0.330.103.418702613.33445.71
@anonymous0.320.132.439376851.371145.50
@nanzo-scoop0.290.074.148591964.38577.23
@eeks0.240.073.5110642314.61435.85
@barrie0.230.063.638322203.78446.02
@tuck-fheman0.200.111.919533502.72765.99
@juneaugoldbuyer0.180.062.9912303034.06546.42
@gavvet0.180.111.666872742.51506.18
@ihashfury0.160.101.578482034.18557.75
@benjojo0.140.131.068702333.73285.44
@mummyimperfect0.110.025.356211544.03456.16
@nabilov0.110.033.2710985452.012119.11

4. Superhumans

They are possibly bots. They have very high N on average (1410.57) and the highest IPR (3.90). They may have very high curation rewards profitability.

AccountInfluencePowerIPRNRangeMeanMaxStDev
@wang1.930.662.9117782138.358210.51
@recursive1.600.315.2211981239.745111.81
@hr11.020.214.80108610410.445212.00
@jl7770.190.053.9024343936.19789.80
@recursive20.180.044.07117310211.508013.88
@taoteh12210.180.044.6710101168.716711.57
@recursive30.170.043.99119510111.838014.98

Conclusion

Curation is a very important task in Steemit. More understanding of curation will provide constructive foundations for further growth of Steemit.

Sort:  

Really cool statistics, thanks!
As we can see Lazy Lions group has the best Avr_Mean of 0.97. So, basically they are the best curators in terms of fair reward distribution among multiple authors. The upcoming changing in SP spending mechanism will harm this group less than others (especially bots) and that's really cool. All the minnows will feel their x8 power too eventually.

Yes, I will post the changes after the next hardfork.

wait so our voting power will increase as well?

I think I would love to see such analysis for users who have between 5000 - 25000 SP

This is a lot of data, and is interesting. Some how it de-humanizes steemit, which is troublesome for me. I know the saying "don't shoot the messenger".

When we splice apart technical data from human curation, I think we're on a bad path, and hopefully this won't be a sign of things to come.

At the end of the day, all of steemit is human interaction, and that includes if there are people behind bots or aliases.

If this is a technological war, I feel that humans will always win in the end. Ever see the movie irobot? (probably). There is a lot of thought and lessons in that movie.

Steemit will never be over-run by irobots in the end. So while you are labelling groups with similarities... this is probably identifying the short-term attributes. It won't last is my prediction. Real thought, by real people, and real participation, by real curators will bleed through all these statistics in the long term.

I hope you are hoping for what I hope for.. because if what I said is true, steemit will win in the long run.

I get your point, however I don't see this post the way you describe it in your comment. In my opinion this is just a very good insight (objectively or not) in how it currently goes down.

I'm not a big fan of "numbers don't lie" hence I totaly agree with your comment but at the same time I have to mention that I see it as a potential wakeup call.

My comment would have made more sense (I now see), if I had quoted the particular section:

Superhumans: They are possibly bots. [...] They may have very high curation rewards profitability.

Looks like I'm fucked, because I'm just a "reader". ^_^

Steemit is more complex to get ahead. Only with the right PR and marketing strategy will do great.

Recursive is definitely not a bot (he participates in conversations), although I don't know if he uses automated ways for his curation.

edit: For curation performance check this out: https://steemit.com/stats/@furion/a-quick-look-at-the-top-curators-and-their-rewards

Recursive around 30%, wang around 24%, hr1 38%.

Possibly both. I found that recursive is mirroring hr1.

Thanks so much for this excellent detailed info

Awesome stuff....thanks for the details and the efforts that you have put in.

Coin Marketplace

STEEM 0.18
TRX 0.13
JST 0.029
BTC 58269.58
ETH 3138.31
USDT 1.00
SBD 2.43