Data Science for the Detection of Emerging Music Styles

When music was still sold on physical carriers such as CDs or LPs, to maximise profits music outlets needed to carefully fill their limited shelf space with the items most likely to bring in the most income, which, assuming equal cost and physical footprint, will be the most popular titles to their clientele. This business model is known as a Blockbuster strategy, and involves heavy investment and promotion of a few select products.

There was some hope that this would change in today's digital economy with minimal overheads for the artists (minimal recording and reproduction costs), retailers (no limits on shelf space, cheap promotion and distribution), and consumers (practically unlimited choice). The expectation was that retail patterns would shift to a business model of selling `less of more', taking the focus away from the elite few and allowing smaller, less well-known artists to prosper. This is known as the theory of the Long Tail (coined by Chris Anderson): while some artists still get the lion's share of the revenue, the tail of less popular music would lengthen and fatten.

Surprisingly, the opposite was found to be true: the tail has become even skinnier, with an even smaller proportion of artists able to make a living from their music. Research by The Harvard business review in 2008 found that 1% of artists account for 32% of total plays on the online radio station Rhapsody, with 10% making up 78% of plays. Similar figures have been quoted by music licensing company PRS music for both illegal peer-to-peer network sharing services and legal downloads, finding for example that 75% of the music stocked by online stores did not find a single buyer.

A well-known explanation for this is given in the book "The Paradox of Choice", where Barry Schwartz observes that having too many options tends to be paralysing instead of liberating. Applied to the popular music market: as searching for new interesting music comes at a cost to consumers (at least an opportunity cost), they will often play it safe to avoid disappointment: they will either listen to the same old bands over and over again, or at best they will try what is recommended to them by trusted parties (friends, or automatic systems that recommend songs liked by people similar to you). As a result, the rich get richer, and revenue concentrates on the hugely popular few.

This makes it increasingly hard for new music trends to gain a foothold in the music industry. Even if a pioneering band's music has a genuine potential of ultimately appealing to large consumer groups, there is only a small chance that it will ever emerge from the skinny tail of popular music. As a result, creative innovation in popular music is stymied, and new emergent music styles disappear before becoming sustainable.

Thus the following question begs an answer: is it possible to detect emergent music styles at an early stage, in a scalable (and thus automated) way, characterising it in terms of its innovative audio features, demographics of the fan base, and their geographical location. Today, for the first time, all stars necessary for doing this are aligned. We have access on a large scale to the audio of a number of bands of the order of a million (e.g. on SoundCloud), and we have access to their fan base and their properties through social media (e.g. Twitter). The subject of this proposal is to gather this data, and to develop the data mining techniques needed to discover new emerging music styles at a very early stage.

This proposal would thus provide the tools necessary for an entirely new way of recommending music that is able to put in the spotlight music that is truly original, currently budding among a small set of fans with a specified demographic and geographical location. Rather than oppressing new trends (as current recommendation strategies do), it would make it possible to actively promote them, and in this way to give new air to creativity.

This proposal has the potential to impact on the following groups:

- The music industry as a whole,
in increasing the efficiency of the music market and the creative economy more generally -- a market that has traditionally been of great relevance to the UK economy and in which the UK has long played a leading role. This will ultimately increase the competitive position of the UK in the international creative economy.

- Music analytics companies (e.g. MusicMetric, uPlaya, SoundOut, Next Big Sound)
can benefit from this research if they integrate the research results into their platforms, in offering a more open-ended way of exploring what is going on in music scene.

- labels and their A&R departments more specifically (major as well as independent labels) as well as companies representing groups of labels (e.g. Merlin Network, The State51 Conspiracy),
in providing them more effective means to search for new market opportunities that would otherwise go unnoticed. This will help them save costs and be more effective at the same time.

- Music players / streamers, music recommendation services (e.g. Spotify, Deezer, We7, Last.fm, iTunes, Nokia Music, etc.),
in making new features possible (e.g. a player / recommendation system recommending music based on a demographic profile and geographical location of the user, instead of just recommending music similar to other music they, their friends, or people similar to them like).

- Companies providing business intelligence and marketing services (e.g. SDL/Alterian, HP/Autonomy, IBM),
in providing an understanding of how demographics relates to music preferences, and this very early on in the lifecycle of a music style. Of course, also for marketing of music specifically our results would be useful. For example, assuming that Liverpool and Manchester have a similar demographic, an emerging music style in Liverpool may well catch on in Manchester as well.

- Music venues, festival organisers,
in giving them original ideas for the organisation of music events and strategically booking bands that will appeal to their clientele.

- Charities and public organisations providing subsidies for music / arts,
in helping them decide where to invest so as to maximise diversity balanced with chances of success.

- Music consumers,
in ensuring a more creative music industry, where new talent is given a fair chance.

- Other vertical markets (besides the music industry) that would benefit from detecting emerging trends from a body of interrelated data, or from finding patterns in relational data more generally.
For example MusicMetric.com is actually a front-end dedicated to the music vertical market of semetric.com, a data analytics company. In a similar way, the data mining results are not limited to this vertical market and will be of use also in other vertical markets. A few examples are:
* Detecting emerging trends in research publications and collaborations.
* Detecting emerging trends in the job market.
* Detecting emerging trends in lifestyle and consumption patterns.
In these examples data will not always be publicly available, but often will be accessible to certain market players.