PageRank on a Graph

Which pages receive more attention when links act as recommendations? Edit directed links, run PageRank, and compare each node’s share. This model describes a random surfer, not current search-engine ranking.

Try it

Enter A -> B. With damping 0.85, B reaches 64.912% and A 35.088%. Add B -> A and both receive 50%. A ring gives equal ranks.

Controls

Each line is a directed link A -> B or a lone node C. IDs contain 1–24 ASCII letters, digits, underscores or hyphens. Duplicate links count once; self-links are allowed. Maximum: 50 nodes and 200 links. Run solves from equal starting shares; Step advances once; Reset restores equal shares without changing the graph. Editing inputs clears stale results. Select a node or its table button to inspect the previous iteration’s contributions.

Model

Start with p(i)=1/n. Update every node simultaneously: p_next(i)=(1-d)/n+d*(sum of p(j)/outdegree(j) over j→i + dangling_mass/n). Pages without outgoing links distribute their share equally. Damping d is the probability of following a link; 1-d is a random jump.

Run stops when the sum of absolute rank changes is at most 1e-10, or after 1000 iterations. The latter is explicitly reported if convergence has not been reached. Percentages show three decimal places, so rounded shares can differ slightly from 100%. All computation stays in your browser.

Small inputs are kept in the existing preference cookie. Larger graphs exceed that storage budget: they remain intact while switching tools in this page, but are not saved across reloads. A short notice appears for these graphs; copy the links before reloading. No graph is silently truncated.