Week 6 gave every page numbers that describe language — TF-IDF weights, topic mixtures, PPMI — and one recurring warning: a cosine without a close reading proves nothing. So we ask one question one character at a time: which pages look closest, are they actually linked, and what do the sentences say? Start on Storm and see for yourself.
Click a row for the shared fingerprint words and the actual sentences behind the match; click a name to make that character the subject. The green ✓ badge is the week-1 Wikipedia link — the one piece of ground truth in the picture.
The same trick, one level down. The subject's TF-IDF fingerprint — their
most distinctive words, names removed — is a row of chips. Click one: the same
word-context matrix, one column weighted by raw counts, one by PPMI. Watch
hammer trade power, to, with for thor, asgard.
Click any pair to put it in the drawer. #1 is the week's favorite: Emma Frost ↔ Jack Frost, pulled together almost entirely by the surname frost. Below it sit the Strikeforce: Morituri teammates — pages that share a comic, its creators and rare vocabulary, and simply never got links between them. The vector suggests; the reader decides which it means.
One frozen run of each representation: change the LDA seed and the topic column reshuffles; our Word2Vec is trained on 727k tokens and its neighbor lists drift between library versions. Every number here is one sample of a method — and the sentence drawer gets you to the evidence, not to the verdict.
Want a wider lens? The same week, less leanness: the similarity cascade, the Marvel map and one-character deep dives.