Content
@
https://opensea.io/collection/dev-21
0 reply
0 recast
2 reactions
woj
@woj.eth
small mvp of the @balajis.eth follow graph idea @rafi took the following data of the top 20% users from a hub and we plugged it into vis.js effect is very clustered for now, if you have ideas on how to visualize it please ping us on gh! play with it here: https://nobuilder.github.io/farcaster-social-graph/
6 replies
3 recasts
15 reactions
balajis
@balajis.eth
Great work! Can you try partitioning the graph into dense subgraphs, like below? This would identify the Farcaster tribes. There are various approximate algorithms for clique detection, dense subgraph identification, and factorization of a matrix into block diagonals that you might use.
4 replies
0 recast
2 reactions
Rafi
@rafi
Thanks! That's the next milestone after extracting user activity graph to do modularity optimization and cliques. It will be fun to see who is the hub for each cluster :)
0 reply
0 recast
1 reaction
woj
@woj.eth
we got a lot of good ideas for improvement yesterday — will iterate on it next week if there are some specific algorithms (eg clustering by replies, location, channel activity) you have in mind please shoot here or in github https://github.com/NoBuilder/farcaster-social-graph
0 reply
0 recast
1 reaction
sahil
@sahil
this is great @woj.eth we’re working on a personalized people and cast recommendation for fc. Using eigentrust to rank profiles. edges and the weights in the graph can be configured.. using follows, likes, comments, recasts. work in progress here https://cast.k3l.io/
0 reply
0 recast
1 reaction
Rafi
@rafi
There is an inspiring 2006 paper theorizing about defending against Sybil attacks based on social interactions. Social network analysis will be an important part of building web of trust in pseudonymous networks you often describe. https://www.math.cmu.edu/~adf/research/SybilGuard.pdf
0 reply
0 recast
0 reaction
