Colin Johnson 💭
@cojo.eth
The #1 rule of building in crypto: ⚔️You will be Sybil attacked After a few instances of possible collusion in our early predictions, we implemented an advanced ML algo that incorporates cluster analysis to protect the integrity of the game. Votes from new users have significantly diminished weights until they’ve proven that they’re voting in a non-collusive manner. We can identify clusters that share similar characteristics and always vote in a statistically anomalous way. We then diminish their votes in weight until they start to trend outside of their cluster. The major benefit to this approach is we don’t have to ban people, and the weighting will never affect earnings if your intuition is right. Here’s a little diagram of how it works.
7 replies
10 recasts
45 reactions
sparkz
@jacy
they prob all voted for tupac
1 reply
1 recast
7 reactions
Matthew Fox 🌐
@matthewfox
Would love to learn more about implementing something like this Any notes/pointers appreciated 🫶🏼
1 reply
0 recast
3 reactions
thumb
@thumb
this is cool, bookmarked. i was wondering how much price plays a factor into this (e.g. 50cent ponder predictions, x dollar warpcast invites) and went down a rabbit hole - sharing here
1 reply
0 recast
1 reaction
SydneyJason
@sydneyjason
How are you doing it with low sample sizes at times - Bayesian priors?
1 reply
0 recast
1 reaction
tldr (tim reilly)
@tldr
100%. Comes with the territory.
0 reply
0 recast
3 reactions
big tone
@bigtone.eth
love this kind of insight. awesome system you developed for this
0 reply
0 recast
3 reactions
thumb
@thumb
cc @compusophy
0 reply
0 recast
2 reactions
