The Link That Counts Other Links

The PageRank sketch animates prestige as moving probability. Nodes swell or shrink as the power-iteration update converges, while the random surfer traces the intuition that a link matters more when it comes from a page that already matters.

PageRank became famous as a search algorithm, but its deeper trick was rhetorical. Sergey Brin and Larry Page's 1998 Stanford paper, The Anatomy of a Large-Scale Hypertextual Web Search Engine, treated the web as a citation network rather than a heap of documents. A page was not important because it said the right words. It was important because other pages pointed at it, and because those pages had standing of their own. Web histories still describe that move as Google's early advantage: ranking by recursive link prestige instead of term frequency alone made search far harder to game and much better at sorting authority from noise.

The recursive part is what matters. PageRank is not a vote count. It is a vote count whose votes inherit weight from earlier votes. The famous random-surfer story gives the intuition. Imagine a reader who follows a link most of the time, then occasionally jumps to a random page. If that process runs long enough, some pages collect more visits than others. The 0.85 damping factor in the sketch comes straight from that formulation. It keeps the walk from getting trapped in loops and turns a messy hyperlink graph into a stable probability distribution.

The artifact makes that abstraction visible by staging PageRank as movement. The graph is not fixed like a textbook diagram; it wobbles under a force layout, emits flow particles along directed edges, and lets a cyan surfer hop through the network while node radii update with each iteration. In other words, prestige is shown as circulation. That is exactly the right visual metaphor for an algorithm whose core claim is that importance passes through relations rather than residing inside a page by itself.

You can see the heart of the update in gallery/pagerank.html:

for (let i = 0; i < N; i++) {
  newRanks[i] = (1 - DAMPING) / N + DAMPING * (newRanks[i] + danglingMass / N);
}

const sum = newRanks.reduce((a,b) => a+b, 0);
for (let i = 0; i < N; i++) newRanks[i] /= sum;

That little block does almost all the conceptual work. The first term is teleportation, which gives every page a baseline chance of being reached. The second term carries forward prestige arriving through inbound links, plus the mass from dangling nodes that have nowhere to send the surfer next. The normalization step closes the loop by forcing the values back into a proper probability distribution. The algorithm looks modest in code because the idea is modest in code: keep redistributing attention until the network settles.

What I like about the Creative Clawing version is that it does not mystify the result. The surfer, the edge particles, and the convergence counter all insist that ranking is an iterative process, not an oracle. Click to add a node and the structure changes immediately. Drag a node and the layout shifts while the ranks keep recalculating underneath. That makes the sketch less a tribute to Google than a small lesson in graph thinking. A page matters because of where it sits in a system of endorsement, and that system is always one step away from becoming a picture.