02805 · Social Graphs and Interactions
Who talks about whom in the Marvel universe?
We mapped every link between 303 Marvel superhero articles. The result is not a map of friendship—it is a map of attention.
Enter the network01
Our question
Does being talked about mean you talk about everyone else?
Each arrow goes from one Wikipedia article to another. Incoming links measure attention received; outgoing links measure how widely an article points outward. Those sound related—but the network says they are different roles.
Most attention received
—
— incoming links, but only — outgoing.
IN
Most outward-looking article
—
— outgoing links and — incoming.
OUT
Across all 303 characters, in- and out-degree have a Pearson correlation of —: related, but far from interchangeable. Article length and editorial style can shape out-degree; cultural prominence can attract in-degree.
02
The degree fingerprint
Most whisper. A few dominate the room.
Choose a degree and switch the axes. On linear axes the long tail is almost invisible; log–log reveals just how far the hubs stretch. Zero-degree nodes are shown at k + 1 on logarithmic axes so the isolates stay in the story. Add logarithmic binning to calm the noisy tail, and a fitted power law to see how far a straight line actually carries.
raw P(k)
logarithmic bins
power-law fit
Each dot is one observed degree value
x = degree k · y = P(k)
Why bin at all?
Out in the tail, most degrees are held by one or two characters, so the raw dots scatter into noise. Binning averages over ranges of k that widen as k grows, which recovers the shape the noise was hiding.
Our scheme
Working in u = k + 1: width-1 bins for u = 1…7, then doubling bins [8, 16), [16, 32), [32, 64), [64, 128). Every count is divided by its bin width, and each wide bin is drawn at the geometric mean of the integers it covers—never the arithmetic midpoint, which would sit wrong on a log axis.
The check that matters
Where the bins are one unit wide, the binned curve must land exactly on the raw dots. Our build script asserts this before the page is generated—turn both series on and look at the low-k end.
Why the fit starts at kmin
A power law is a claim about the tail, not the whole distribution. The low-degree end is shaped by how thoroughly each article happens to be written, and forcing a straight line through it would drag the exponent toward whatever the bulk does. So we scan candidate cut-offs and keep the one whose fitted tail matches the data best, measured by KS distance.
What we did not do
We never fit a slope to the binned points. The exponent is a maximum-likelihood fit to the raw degrees above kmin. And a fitted line is not evidence: it will happily fit data that is not a power law at all. Whether this one really is one is a week-2 question.
03
Explore all 303
A living map of attention.
Drag, zoom, search, or select a node. Size encodes undirected degree; colour encodes a detected community. Position comes from a force simulation—it helps reveal structure, but it is decoration, not data.
Lines show links without arrowheads for legibility. Direction is preserved in the selected node’s statistics.
04
Beyond the giant component
Nine heroes inhabit a universe of their own.
The giant component contains 277 characters. Outside it sits a connected nine-character island from Strikeforce: Morituri, plus 17 isolates. A category can put heroes in the same roster without their articles ever linking to the wider network.
91%
of all nodes live in one giant component
What surprised us
Fame and narration are not the same thing. Spider-Man is the network’s destination; Betsy Braddock’s article is its guide.
05
Method & receipts
What exactly did we measure?
Representation
A node is a Marvel superhero Wikipedia article. A directed, unweighted edge A → B exists when A’s running article text links to B. Redirects are resolved; navbox links are excluded.
Construction
We added all 303 roster entries before the 1,784 edges, retaining 17 isolates. For components and density we collapsed reciprocal arrows into 1,434 undirected pairs.
Sanity checks
The script asserts n = 303, directed m = 1,784, undirected m = 1,434, and 17 isolates against the course data page before generating this site.
Human check — do this together
Open Spider-Man’s row in the edge list and count a small sample of incoming links by hand. Record who checked it and whether the count agreed before sharing the site.
Frozen snapshot: 26 August 2026 · Source: 02805 shared playground · Analysis: source repository