"Colourtext always take a fresh approach to discovering hidden patterns in data. Whether you want to conduct segmentation or mine your CRM for deep insights, they will help you see your customers , and what they are seeing , in new ways."
Using network based segmentation techniques to bring a little order to bar room debates about music genre classifications
Colourtext has spent years collecting music genre classifications developed by various authoritative sources in the music industry. They're all very cool, but they're also very different.
So, we decided to build a network-based algorithmic solution to unify these classifications into a new map of popular music culture. You can download a free hi-res copy of the map via the button at the top of this page. If you'd prefer a copy with a white background drop us an email and we'll send one directly to you.
We used a machine-led approach to aligning and merging music genres tags, which was an interesting approach to take because genre labels are often applied to songs and artists in a very human and subjective way.
We started out with around 400 manually assigned genre labels derived from 6 independent music genre classifications that we judged to be the best in our collection. This rich, but rather chaotic and often contradictory, diversity of musical opinion was then rendered down into 27 essential 'meta-genre' music tags by our algorithmic approach.
One interesting feature of the analysis was that on a song-by-song basis more than 90% of artists were allocated to a consistent music meta-genre on 80% of occasions.
For instance, Kings of Leon are genre-tagged 64 times in the database and on 54 occasions (84%) they were tagged as 'Alternative/Indie'. Most artists are 'emphatically' regarded as belonging to one dominant genre, as is the case here.
For a minority of artists the picture is slightly (but only slightly) less clear cut. To illustrate, OutKast are genre-tagged 40 times in the database. On 21 occasions (53%) this band was tagged as 'Hip Hop/Rap' followed by a handful of songs classed as either 'Urban', 'R&B' or 'Pop'.
If nothing else, this analysis tells us we're on fairly safe ground when applying dominant genre tags to individual artists. There doesn't seem to be much benefit in a finer grained effort to apply genre tags at the level of individual songs.
For a little more insight here's a 4 minute clip from a GAS Podcast video interview with Colourtext founder Jason Brownlee discussing the background to the music genre map and how it was made.