Definition
Plain language
The tendency of connected things to be alike — papers mostly citing papers in their own field, friends sharing tastes.
As stated in the literature
A graph property measuring how often linked nodes share the same label; high homophily makes neighbor-based prediction strong and low homophily makes it backfire, determining when a trivial neighbor-lookup beats a trained graph model.
Why it matters: It determines whether predicting from neighbors is a reliable shortcut or a trap, since the same trick that helps in like-with-like networks backfires when neighbors are usually different.
For example, a paper on cell biology mostly cites other biology papers, so guessing its topic from its neighbors works well.
Heard on the show
“Whether "ask your neighbors" is a good idea depends on a property of the graph called homophily.”Episode 144 — When an AI Agent Just Copies Its Tool — And Bigger Models Copy More