A survey is sent to car buyers with a huge number of questions related to their satisfaction. Automakers, Tier-1 suppliers, and probably others get to buy this data to mine for market research.
At a past role with a Tier-1, we bought part of the dataset. A quick regression showed that satisfaction was strongly correlated with buyer age, and there was very little signal otherwise. (Young people with overtime work and daycare pickups don't respond to surveys unless they have a serious axe to grind while retirees aren't so constrained)
Skimming the paper: it looks like they are performing sentiment analysis on books, effectively taking the Fourier transform of the sentiment-versus-time data, and reporting which Fourier component (up to 3rd harmonic) is strongest.
At a past role with a Tier-1, we bought part of the dataset. A quick regression showed that satisfaction was strongly correlated with buyer age, and there was very little signal otherwise. (Young people with overtime work and daycare pickups don't respond to surveys unless they have a serious axe to grind while retirees aren't so constrained)