Similar Artists Finder

How the Similar Artists Finder Works: Last.fm's Collaborative Filtering Explained

Last.fm has recorded over 100 billion scrobbles since 2002, making it one of the largest music listening databases in existence. Every scrobble is a listening event: a timestamped record of which track played, on which account, in what sequence. From this stream of data, Last.fm builds a similarity graph that connects artists based on shared audiences.

Listener Overlap: How Artists Become "Similar"

The core mechanism is listener overlap. If 80% of people who regularly listen to Boards of Canada also regularly listen to Aphex Twin, those two artists score high similarity. The score is not based on genre tags or audio analysis. It is based on real listening behavior across millions of accounts.

This makes the system powerful for cross-genre discovery. Artists that share audiences but differ in sound still cluster together. Nick Cave and Tom Waits appear similar not because they sound identical, but because fans of dark Americana and literary rock tend to listen to both. Genre tags alone would not surface this connection reliably.

According to Last.fm's artist.getSimilar API documentation, similarity scores are calculated from the listening patterns of Last.fm users and updated on a rolling basis. The API returns a match value between 0 and 1. This tool displays that as a percentage.

The Scale of the Data

Last.fm crossed 100 billion scrobbles in 2018, and continues to grow as users connect Spotify, Apple Music, and other platforms via official integrations. The platform indexes data from over 300 million registered users, with active listeners spanning more than 10 million unique artists. This scale gives the similarity algorithm enough data to surface meaningful relationships even for moderately niche acts.

For very small artists, the data may be too sparse to produce reliable similarity scores. If an artist has fewer than a few thousand total scrobbles, the algorithm has little to work with. In those cases, try a slightly more well-known artist in the same niche as your starting point.

The "Most Obscure" Sort Mode

Sorting by "Most Obscure" re-ranks results by listener count, lowest first. This surfaces the least-known artists who still share your seed artist's audience. It is not a perfect measure of obscurity. Listener counts reflect how many Last.fm users have scrobbled an artist, not total streams or record sales. But it is a practical proxy.

An artist with 5,000 Last.fm listeners is genuinely niche. An artist with 2 million is well-established. If you want to find the underground acts that sit adjacent to a mainstream artist, this sort mode digs them out. For dedicated obscure artist discovery, use our Obscure Music Finder, which adds a listener count cap filter directly.

Practical Use Cases

Rabbit-holing into new artists: Start with an artist you know well. Find their top similar acts. Pick the one you least recognize and search again. Repeat. Within three or four jumps you can reach genuinely unfamiliar territory while staying connected to music you already enjoy.

Finding support act candidates: Promoters and booking agents use similar-artist data to identify acts that share an audience with a headliner. High match score plus low listener count often signals a promising support act: same fanbase, not yet widely known.

A&R research: Labels looking for the next breakout in a genre can use this tool to map who the established acts' audiences already listen to. A high-match, low-listener artist who appears in multiple similar searches is a strong candidate worth watching.

Playlist curation: Building a coherent playlist is easier when you start from listener behavior data rather than genre tags alone. High-match artists tend to sit together well in a listening session because their fans already experience them that way.

Limitations of Collaborative Filtering

Collaborative filtering reflects the listening habits of Last.fm users specifically. That audience skews toward certain demographics: historically tech-savvy, English-speaking, and interested in rock, electronic, and indie music. Regional pop scenes, non-English language music, and very recent artists may be underrepresented.

A 2014 analysis by Roskilde University examining streaming data found that the top 1% of artists captured 70% of streams, with the bottom 95% accounting for only 10%. This concentration means the similarity graph has dense, high-confidence data at the top and sparse data at the edges. The further you go from mainstream acts, the less reliable individual similarity scores become.

For track-level similarity rather than artist-level, use our Similar Songs Finder. For mapping relationships between multiple artists visually, try the Music Artists Map.

How to Read the Match Score

A match score above 60% indicates a very strong listener overlap. These artists genuinely share a core audience. Scores between 30-60% mean significant overlap but more variation in listener habits. Below 30% suggests a loose connection: some fans listen to both, but the audiences are fairly distinct.

Do not treat low scores as "bad." A 15% match artist might be exactly the right amount of different for a transition in a playlist. And for discovery purposes, a 20% match artist from a genre you have never explored can open a new category entirely.