Example Use Case

Farcaster Follower-based Scores

We applied EigenTrust to Farcaster, with the goal of getting a ranked list of people to follow. The intuition is that, if A follows B, A would be interested in following who B follows.

In this experiment, the local trust matrix consists of followership: Cij = 1 if i follows j, 0 otherwise. (We did not introduce different levels of trust among followership, such as family/friends/acquaintances, as Farcaster did not have those categories.)

We set the seed peer to be the peer themself for whom to calculate the ranking. In this model, random celebrities did not count unless they were in my follower-based social graph. We calculated 4 sets of trust score rankings, using the seed confidence of 0%, 50%, 80%, and 100%.

With the 0% seed confidence, we saw that the top ranked people were almost entirely network-wide celebrities. As we went from 0% to 50% to 80% confidence level, other people started floating toward the top of the ranking: Those followed more in the seed peer’s vicinity on the social graph. Finally, at 100%, the ranking consisted solely of seed peer; others received 0 scores: This extreme, although not useful, demonstrates how seed confidence works.