A content affiliation graph is a weighted graph G = (V, E, w) where each node v ∈ V is a content cluster, and an edge (u,v) ∈ E exists when you co-read topics u and v within a session window τ.
A structural hole is a gap between two clusters that are not directly connected in your map. Burt (1992) showed that bridging structural holes is where the most novel information comes from.
The simplest measure: how many other clusters does this one directly connect to? In your map, a connection exists when you regularly co-read two topics in the same session.
How few hops does it take for information from anywhere in your map to reach a given cluster? Short average path = central cluster. Long path = peripheral, isolated cluster.
How often does a cluster appear on the shortest path between two other clusters? High betweenness = bottleneck. This is the metric most directly tied to echo chambers — all information passing through one node is a textbook filter bubble.
Not just how many connections, but how well-connected those connections are. Like Google PageRank — a link from a high-authority source counts more than ten from isolated echo chambers. Below 0.5 often signals an echo cluster.
Like eigenvector centrality but counts all paths — including long indirect ones — with a penalty (α) that shrinks with distance. This is why Fairxis can recommend content 3 hops away and still explain its relevance to you.