Dynamic Clusters – Maybe You CAN Get There From Here

Fun thing about blogging is the participation of commenters: already got some good hints in comments on how dynamic peer clusters may form.

I previously posted that “dynamic peer clusters” – groups of people who like the same books or movies or music or whatever as you do – are a perfect source for recommendations of new books or movies or music or whatever.  You are all scouts for each other and you all have roughly the same tastes – especially in whatever.  But, I lamented, what incentive do people have to provide the information necessary to establish the clusters in the first place? The clusters don’t become useful – they don’t give anything back – until a lot of people have done some work to put information in.

In a comment, reader vcmc cites last.fm: “I'm finding it's performing pretty well for me as a dynamic peer cluster, via the ‘neighbors’ the system generates for you after a week or two of tracking your listening patterns.”

last.fm describes musical neighbors as “people that the system automatically detects have a similar music taste to you.” last.fm learns your music taste by noting what tracks you listen to and how long you listen to them.  It also has a downloadable client called an “Audioscrobbler” which you can run on your machine to let it know what music you listen to from other sources like iTunes.  You can listen to a customized “radio” station based on your musical profile.  More to the point of dynamic clusters, you can also listen to virtual stations based on the preferences of your musical neighbors.  Looks to me like you can get to meet your musical neighbors as well if they want to be met.  You can certainly see their profiles.

OK. last.fm has figured out how to get there (a critical mass of people to make dynamic neighborhoods) starting from here (not enough users for any social value).  They have established value from the first user on by offering free music to listen to.  They give their users a reason to compromise their privacy by running “Audioscrobbler” because the more last.fm knows about you, the better job it can do in customizing your radio station.

They hold out the promise of making new friends who have similar tastes.  They also offer artists a free way to promote music which, in turn, makes more music available for downloading.

(note: I may be the only unmusical person in the blogosphere so can’t give any useful input on how good last.fm really is.  I just roamed the website for awhile after I got vcmc’s recommendation and scrolled through a pretty good guided tour.  My interest is in effective ways to get people to arrange themselves in dynamic peer clusters and this may be one.)

Reader vcmc does have one criticism: “…the assumption is made that the more you listen to a track, the more you enjoy it. Which - in general - is accurate, but there are always exceptions. I always feel like I'm ‘poisoning’ my account if I listen to a new record a few times to see if it sinks in but I don't end up enjoying it.” vcmc suggests a mashup of last.fm and metacritic to get around this problem.

Which brings us to a comment by Jason about Netflix which does solicit customer ranking.  “Every time I rank a movie, their recommendations get better. So I keep ranking movies. That's one way to encourage the user to deposit data.”  So Netflix may be building clusters and using them to generate recommendations.  Mary and I stopped using Netflix because we were having a hard time thinking up which movies we wanted to see.  If we’d been more patient, perhaps we would have gotten recommendations which helped us over this hurdle.

Jason asks: “What if one company aggregated data from Netflix, Amazon, Last.fm, Pandora, etc., etc., and then both re-supplied the larger data sets to each collector and used the database to create their own recommendation sites.

 

“In that case, I guess ‘getting here to there’ would involve someone persuasive enough to talk the big data collectors into selling their data.”  There is also the problem of matching up the different identities each of us have with different services and there are sure to be privacy concerns.  Still, it’s an intriguing idea.  There is probably a pretty strong correlation between what books people like, for example, and what movies.  Don’t know about music, of course.

Again, Netflix has found a way to give a user an almost immediate advantage – better suggestions -  if the user takes the trouble to rank movies.  This starts out being just an added benefit of a service which people love because it eliminates late fees. But, in the end, when there are multiple good ways to get video on demand, users will resist switching from a service which includes their neighborhood, their dynamic peer cluster.

Amazon understand that, of course, but seems to have been almost reticent to do more than the simple correlation of people who bought this book also bought this one.

It may be that dynamic peer clusters become a sine qua non for online retailers, particularly in the arts.  The retailers do have the purchase or listen or view or download information.  They don’t have the feedback unless they solicit it but Netflix has shown that it should be easy to get users to provide this feedback.  It may be that the business opportunity is in providing retailers with the tools to do the clustering well (if you’re not a retailer yourself).  Certainly this is AN opportunity.

But I think there’s more opportunity than just enabling retailers – as big as that opportunity is.  Look forward to hearing more good ideas from you (or seeing the alpha of your new dynamic peer clustering service – I’m sure you’ll have a better name for it than that).