When I reviewed MusicStrands at the end of last year, I noted something odd about the recommendations that the system gave me. I started entering a playlist that I’d already entered on several other similar services (including Art of the Mix, Mixmatcher, FIQL and GoFish). When I was half-way through entering the playlist on MusicStrands, I noticed the recommendations that MusicStrands was suggesting were exactly the same tracks that made up the second half of the playlist, as published elsewhere.
As this could not be a coincidence, I posted a MusicStrands journal entry in January to ask if MusicStrands was importing data. When I got no reply to this, I sent a message in February to Byron Prong, described on his profile as “the resident Musicologist and helpful guide to the MusicStrands site”, referring him to the journal entry. Still no reply or acknowledgement at the time of writing this.
Mike Wu of FIQL has assured me that he hasn’t licensed any of his playlist data to MusicStrands. There’s nothing wrong in principle with one service provider making such data available to others to generate recommendations, as long as no personal data is involved and no privacy is infringed. I’m not sure if there would be any way for one provider to ‘harvest’ another’s playlist data without their permission. So I’m not levelling any accusations at MusicStrands, but you’d think that, if there were nothing to be embarrassed about, I might have got a reply by now.
Update, 5 April 02006: I’ve now had an email from MusicStrands, which I have permission to quote.
MusicStrands uses artificial intelligence and advanced data mining techniques to ferret out and process relationships among items of interest (songs, artists, albums, etc.) extracted from playlists. We not only learn about these relationships from playlists users upload to MusicStrands, but also from playstreams played through our MyStrands client, and even from the “skips” users generate. Be assured that no identifiable personal data is collected or stored during this process of building what we call our “ground-truth”… This ever evolving ground-truth, combined with the specific taste and context of each individual making a request, is what
allows MusicStrands to provide its personalized recommendations.
With that background, and taking into account the astounding nature of your finding, we believe there are at least two possibilities. First, it could have happened that one of our users uploaded your playlist from an external resource to his/her iTunes or WMP [Windows Media Player], and played it through MyStrands themselves, in that case the playstream would have been analysed and the relationships have become part of the ground-truth. A second alternative that we initially considered, was the possibility that one or more of our members playstreams together contained all of the songs in your playlist.
As their email goes on to say, the probability of the second explanation seems infinitesimally small. They also pointed me to the fascinating MusicStrands labs mini-site, which I’ll report on separately.