Pic via agiledogs
One of the things that frustrates me about recomemndation engines, but particulaly is how bad they can be. How the slightly off-kilter recommendations are magnified a thousand times.  Part of this magnification is I believe because of the fact that we’ve become atuned to believing that the software is somehow “automagic”.

Whenever I speak to people about the general feedback is “yeah, but it’s quite patchy”. The algorithm is always to blame, or the underlying music data (and publishers they’ve got on board) but rarely the user.  During a conversation last week with Tom it struck me that the experience of presenting relevant music should feel far more involved than it is, in other words the emphasis on training the software to perform better for you should be more central to the experience (and not reserved for technical criticism). You get the music you deserve.  This has all sorts of potential gains too.  I mean you can start to record how active you are at training and report back how well the dog software is performing by seeing how many times you *love* a track and how often you skip.  in order to make this work you need to show what success looks like.  If you go to dog training you know you get a good dog, a dog that behaves (i.e. it does what is expected and / or what you want).  For you could to present back:

1. “just discovered <blah track> ” -> i.e. trial -> recommendation

2. purchases in the last <time period>. Purchases being a good proxy for how well the service is delivering trial -> conversion

The current recommendation system doesn’t work effectively because it’s based on an un-selfish act.  The act of recommendation is conciously, thoughtfully delievered and potentially comes with lots of baggage like wanting to be seen to be cool.

And we can perhaps learn another thing from training dogs.  People take puppies to be trained.  However, a lot of people take dogs that need remedial training.  Lazy dogs (and even lazier owners) who need a quick fix. could perhaps provide that quick fix.  This could be a ‘remedial’ training of the software.  I can’t be arsed to play 100 Fall tracks to weight my music playback in a certain direction.  And I want to ban mediocre guitar bands for ever, so no Elbow please. You know, quick fixes. This then represents a scale of emotions and different user neeeds, but more importantly it puts the onus back onto the user which as a piece of social software makes the service potentially all the more interesting.