Two hundred thousand HuffPost headlines record a digital newsroom mid-transformation: politics triples its share of the front page, the listicle rises and dies, and every word β from zika to #MeToo β arrives on a date you can point to.
A newsroom's category labels are its org chart made public: every headline gets filed somewhere, and where it gets filed tells you what the institution thinks it is. HuffPost's archive, 2012 to 2022, happens to preserve those labels through the most disruptive stretch in modern media β the Trump ascent, the pivot away from lifestyle verticals, the death of the traffic-bait formats that built the site.
In 2014, politics was one desk among many at HuffPost β under 10% of output, outnumbered by Wellness and Entertainment. Then came the 2016 cycle. By election month, 38% of everything the newsroom published was filed under Politics, and the lifestyle verticals that once defined the site β Style, Taste, Weddings, Divorce β were flattening toward the margins. The chart below shows the whole compositional shift; the red line is election day.
Formats have life cycles too. The listicle β "7 Waysβ¦", "23 Photosβ¦" β peaks in 2014 at 4.5% of all headlines and then falls off a cliff, down to half a percent by 2018. Question headlines ("Is Snapchat My New Nemesis?") die alongside it, from ~9% to under 2%. And as the gimmicks retire, headlines get longer: the median grows from about 9 words to 11, the shape of a newsroom writing for search and social-preview cards instead of the slot-machine curiosity click.
Finally, the archive works as a linguistic seismograph. For 25 tracked terms, the chart below marks the day each first appeared in a HuffPost headline. Some arrivals are pandemic-grade events (zika, January 2016; brexit, February 2016 β four months before the referendum). Some are slow cultural creep (selfie, 2013). And one is a trap for the unwary text-miner: coronavirus first appears in September 2012, in coverage of MERS β seven years before COVID-19.
Source: News Category Dataset v3 (Rishabh Misra), 209,527 HuffPost headlines with categories and dates, 2012-01-28 β 2022-09-23. Listicles matched by pattern ("N things/ways/reasons/β¦"); question headlines contain "?"; term arrivals are word-boundary regex matches on lowercased headlines. Figure 1 groups the 9 largest categories; the rest are "Other."
The archive is dense 2012 β May 2018 (~200K headlines) and sparse afterward (~10K for 2018β2022): trend claims are scoped to 2012β2017 throughout. The POLITICS label exists only from 2014 β earlier political coverage sat in other verticals. Term first-appearances are literal string arrivals (isis matches the model Isis King before 2014).
Data: CC BY 4.0 β attribute Rishabh Misra, arXiv:2209.11429. Headlines Β© HuffPost. Article text and figures: CC BY 4.0.