We were in part inspired by Kurt Vonnegut's Shapes of Stories:
The timeseries are computed as a rolling average for a sliding window of 10,000 words, and sliding forward in increments of 1,000 words. The comparison and reference selectors compare the percentage of text which they are above. Note that for shorter books, the sliding window leaves noticeable room at the ends, but that text can still be compared. The lens slider allows omitting a group of words from the tool, with the default for books being a stop-window of 3 through 7. To see only positive words, or only negative words, slide the window to the ends of the distribution.
For more information about wordshifts, see http://hedonometer.org/shifts.html.
To download the word vectors used to make the visualizations, we provide the data as a .tar, .tgz, or .zip: