A word that appears a lot in this document but rarely across all documents is probably distinctive. TF-IDF scores it high. Words that appear everywhere score low.
TF-IDF(t, d) = TF(t, d) × log(N / DF(t)) · where TF = count in d, DF = #docs containing t, N = total docs
CORPUS (edit any document)
Reading the matrix: "the" appears in every doc → DF = N → IDF = 0 → TF-IDF = 0. Useless for distinguishing docs. A word that appears only in Doc 2 has high IDF and a big TF-IDF score in Doc 2 — flagging it as that doc's signature term.