what gets played back

152,090 records with a real play count, by what they talk about.
the number, first

every “plays” figure on this wall is a last.fm scrobble count: every time one of its users’ players reported playing the record, since 2002, plus how many different people did the reporting. it is a real count of real plays — the only one any licensed, free source publishes. spotify and apple publish none; deezer publishes an index of how hard a track streams today, which is the “still streaming” chip, and the two are never mixed.

it is also a sample. last.fm’s users are the people who bother to scrobble — album listeners, crate people, a crowd that leans older and more inward than the streaming charts. so a 1996 album cut can outplay a 2023 single here, and everything on this page should be read as what that crowd replays, not what the world streams.

a number was kept only when the artist and the title both matched what was asked; last.fm’s autocorrect likes to hand back a different record with a similar name, and a wrong number on a real record is worse than none.

NO EDGE
how it sits in the graph

there is no line between plays and themes. plays are a fact about the song — Song.lastfm_plays, Song.lastfm_listeners — and a theme is an edge the song carries, HAS_TOPIC, written when the lyric matched the theme’s lexicon. they meet at the record and nowhere else, so every figure below is one hop: take the songs that carry a theme, read their plays, take the middle one.

MATCH (s:Song)-[:HAS_TOPIC]->(t:Topic)
WHERE s.lastfm_plays > 0
RETURN t.name, count(s) AS songs, avg(s.lastfm_plays) AS plays
ORDER BY plays DESC

medians below, not averages: a handful of forty-million-play records would otherwise decide every theme. the full shape of the graph is in the ontology.

every theme, by the plays of its records
themes with at least 300 records that carry a play count. the bar is the median; under each name, the single most played record about it.
hits are about one thing
median plays by how many themes a record carries
the more themes a lyric reaches for, the fewer times it gets played back. the records people return to tend to be about one thing; the dense, everything-at-once records are the deep cuts.
the words effect
median plays, split by whether the wall has the lyric
this one is not about themes at all. a record with its words on file is a record somebody transcribed, which is a record people listened to; the ones without are the deep catalogue. so “tagged records play more” is true and uninteresting — the comparisons that mean something are the ones above, among records that all have words.
the 2010s alone
the same ordering inside one decade, so it is not the old catalogue doing the work
READ CAREFULLY
what this does and does not say
  • it says last.fm’s listeners replay the inward records — heartbreak, pain, apology, school, freedom — about three times as often as the street records: the trap, the strip club, pimpin’, haters, the come-up.
  • it does not say rap listeners prefer them. the sample leans toward album people; the streaming charts do not.
  • volume drags the street medians down. the trap & the block is tagged on more than twelve thousand records, most of them mixtape-era deep cuts. the median is honest about that; it is not a verdict on the theme.
  • love includes the singers. the wall carries r&b, and r&b gets played. love and heartbreak would sit lower on a rap-only wall; the ordering past them holds either way.
  • a theme needs 300 played records to appear here; counts are medians; every figure is as of the fetch date above.