Definition
Plain language
The classic web-search idea of ranking things by how they're connected, by imagining a random surfer wandering the links.
As stated in the literature
A graph-centrality algorithm based on the stationary distribution of a random walk over links; the personalized variant biases the walk back toward seed nodes, used in ExpGraph to diffuse from query-similar memory nodes to indirectly related ones.
Also called: personalized PageRank
Why it matters: Its idea of ranking by connections lets systems spread relevance from one known item to related ones, useful for both web search and retrieving related memories.
For example, it imagines someone randomly clicking links forever and ranks a web page higher the more often that wanderer would land on it.
Heard on the show
“The technical engine is personalized PageRank, which is the same random-walk idea behind early web search, just pointed at this memory graph instead of the web.”Episode 106 — Giving Agents a Notebook Instead of New Weights: How ExpGraph Lets Frozen Models Learn