The dataset
Everything MusicRoots holds is available as a file: 1055 artists, 2534 connections between them, and 2656 citations behind those connections.
The licence is CC BY-SA 4.0. Use it for anything, including commercially, as long as MusicRoots is credited and anything you redistribute carries the same licence.
Download
- musicroots-influences.json — 1.6 MB, 2534 rows. Everything: artists, connections and citations, in one file.
- musicroots-artists.csv — 141 KB, 1055 rows. One row per artist.
- musicroots-influences.csv — 373 KB, 2534 rows. One row per connection. Read as: artist was influenced by influenced_by.
- musicroots-sources.csv — 550 KB, 2656 rows. One row per citation, joined to a connection by the first two columns.
What is in it
Two things, and the second is the unusual one. The connections — artist A influenced artist B — and, for every single one, the published source that says so: an interview, a memoir, an encyclopedia, a documentary, an official publication. A connection with no source is not in the database, so there are none in the file either.
What is deliberately absent: anything derived from audio similarity, listening data, genre tags or chart position. If two artists sound alike and no source links them, no row here joins them. What MusicRoots is sets out the method in full.
How the tables join
musicroots-influences.csv has one row per connection, with artist and influenced_by both holding slugs. Read a row as: artist was influenced by influenced_by.
musicroots-sources.csv has one row per citation and joins back on those same two columns. musicroots-artists.csv is keyed on slug, which is also the last part of that card's address on the site, so any row can be checked against the page it came from.
The JSON file carries the same data nested, plus a meta block naming the licence, the version and the counts.
How to cite it
MusicRoots, Influence dataset, version 2026-08-14. https://music-roots.org/data — CC BY-SA 4.0.
The repository also carries a CITATION.cff, so GitHub will format a citation for you.
Corrections
The file and the site are the same data, so a correction to one is a correction to both. If a connection looks wrong, or rests on a source that does not support it, report it — it takes a URL and nothing else.