Pick a page. Its textual lookalikes are the pages whose TF-IDF vectors point the same way. Every word in that vector is a name, a habit word (she, her, voiced, series: stopwords and words on a third of all pages) or something else. Turn the first two dials down and watch the list re-rank, and what the network makes of it.
100: every name at full weight. 0: names removed (the vectors are renormalised).
100: stopwords and words on a third of the pages at full weight. 0: removed.
The curve is the average over all 303 pages as the names dial goes from 0 to 100, at the habit dial's current setting.
What is being turned. Each page's TF-IDF vector (tf = count / words, idf = ln(303 / pages with the word), unit length) is the sum of three pieces: its names, its habit words and the rest. A dial multiplies one piece by its setting, and the vector is renormalised. The cosine of two pages is then a sum of the three parts shown in the bars, and they add up to the number on the right. At 100 and 100 this is the course's TF-IDF; at 0 and 100 it is TF-IDF with the names removed; at 0 and 0 names and habit words are both gone. linked means a Wikipedia link in either direction between the two pages in the week 1 snapshot.
Names: words capitalised in the middle of sentences, or in a page title. Habit words: NLTK stopwords, and words on at least 101 of the 303 pages. The post audits both rules. Click a name in the list to move to that page. The state is in the address, so a link keeps your dials.