GnuMusiq/details

GnuMusiq starts with uploads of CC-licensed audio files, to build up a library of material that is freely available for listening or download.

One of the main problems with large media libraries, whether free (e.g. YouTube) or for-purchase (e.g. Google Play) is finding material that you like. Most sites solve this with rather crude algorithms, e.g. in Pandora Radio the user selects a particular artist, and the algorithm finds music that is liked by other users who also like that band. "Liking" is up/down binary, with no gradations or categories.

GnuMusiq would allow users to create categories, and rank music on a scale of -10 to +10 within that category. For instance: I might really like both Led Zeppelin and Enya, but someone else who likes Zeppelin might not want to hear Enya (or vice-versa). Category ranking would allow me to create "heavy metal" and "new age" categories, into which I could highly rate Zeppelin and Enya respectively. Any track could be rated into as many categories as the user cares to bother with (e.g. any Pink Floyd track would probably rate highly as "classic rock", while only some would rate as "ambient" and others would rate as "heavy metal" or "electronica").

As users build up their personal ratings-libraries, it becomes possible for other users to make use of them:

• If I find a user whose taste I generally agree with, I might use one or more of their categories to build a set-list, which could then either be streamed (as in the Pandora Radio model) or downloaded to a device for listening off-line (not possible with non-libre material).
• The system could cross-correlate ratings-libraries of different users, looking for topics with a lot of overlap but some differences, and use this to build a set of "consensus categories" containing many more tracks than any one user would have time or inclination to rate. These consensus categories could then be promoted to new users as a good way of getting started.
• Users could also rate the category-collections of other users, giving some sense of their overall quality (and of the overall quality of each user's curation skills), which could then be used to produce lists of recommendations for other users.

Since this level of rating is a bit more work than simple up-down voting, we will probably want to provide some incentives and rewards for users who do take the time. This could include store credit (for buying artist merch, artist patronage, physical media, etc.). The incentive system needs to be carefully designed in order to avoid creating "perverse incentives", but this doesn't seem like a huge obstacle given that we will be getting a lot of feedback from a large number of users, even if that is only a percentage of the total audience. It should be easy enough to spot outliers who, for example, produce an unusally high volume of ratings and yet whose categories do not correlate well with anything and who are not themselves highly rated.