Retailers such as Amazon and Half use social network methods applied to people’s previous purchasing behavior and demonstrated interests to figure out what other items customers may want to buy. MovieLens is an interesting example of a non-commercial service that uses information provided by the user about his or her movie preferences (ratings of movies already viewed) to suggest what additional movies may be of interest to the person based on the movie evaluations of others who exhibit similar tastes. Music Plasma suggests what artists are close to each other based on style and epoch. Unfortunately the site doesn’t tell us much about the underlying methodology.[1] Unlike MovieLens, it seems to rely on information about the position of artists in the network based on shared genre and era to make recommendations (i.e. display linkages) instead of relying on listener feedback about shared tastes. I’d be curious to hear about other similar services resembling any of these approaches. For those interested in visualizations of this type, the search engine Kartoo and the Virtual Thesaurus may also be of interest (the latter is quite restricted for non-subscribers though and I have never been able to access enough of it to be particularly impressed). For more on visualization of networks see orgnet.com.
fn1. A few months ago I contacted them for more information, but got no response.
{ 1 trackback }
{ 14 comments }
John Emerson 02.07.05 at 1:04 pm
The Music Plasma designer was an ENORMOUS admirer of Warren Zevon.
John Emerson 02.07.05 at 1:36 pm
Google Mozart and you get Raffi, Sesame Street, and Music for Little People. Not kidding.
x 02.07.05 at 2:06 pm
http://www.musicmobs.com/
http://www.audioscrobbler.com/
Those are based entirely on members preferences. You can find related artists, popularity lists, etc.
John Emerson 02.07.05 at 3:16 pm
Hm, I just used “Google” to mean “search” (in Music Plasma). Talk about a violation of IP rights!
Des von Bladet 02.07.05 at 3:38 pm
I definitely remember a social networking service in the cool Jurassic dawn of the Interweb bubble, but Google refuses to confirm my vague memory that it was called “Firefly”.
But in any case it sucked quite badly: the decisive advantage of Amazon’s implementation for me is precisely that it isn’t the feature attraction. If it’s useful, it’s a bonus; if it isn’t it’s no great loss.
Even at its best, social networking loses out to a really good Amazon list or other annotated thingography, in my experience.
Simra 02.07.05 at 6:47 pm
A similar site is http://recommendz.com. That one is based on research by Matt Garden and Greg Dudek. Their FAQ describes a little bit of the science behind collaborative filtering. Disclaimer: Greg Dudek was my PhD supervisor. :-)
R
Simra 02.07.05 at 6:48 pm
A similar site is http://recommendz.com. That one is based on research by Matt Garden and Greg Dudek. Their FAQ describes a little bit of the science behind collaborative filtering. Disclaimer: Greg Dudek was my PhD supervisor. :-)
R
Simra 02.07.05 at 7:02 pm
D’oh! Sorry for the DP.
Brian 02.07.05 at 8:53 pm
Des von Bladet:
IIRC, firefly came out of MIT. I don’t remember being particularly impressed by it, though.
Leonard Richardson 02.07.05 at 11:11 pm
As luck would have it I just released a recommendation engine for web pages which gives recommendations based on the contents of the weblogs you read and write. Since it treats a web page as a “user” who recommends links, you can use it without getting a lot of real users to give ratings first.
simra, thanks for the pointer to Recommendz.
Andrew Cholakian 02.08.05 at 1:14 am
There’s also the excellent http://www.whattorent.com/
Peter Levine 02.08.05 at 1:14 am
As a tool for mapping the link structure of portions of the Web, I like TouchGraph (http://www.touchgraph.com/index.html) better than orgnet, although both have irritating drawbacks.
Tom Davies 02.08.05 at 11:07 am
A few months ago, Amazon recommended the 9/11 Report to me. I assumed that this was because I’d bought Christopher Hitchens’s ‘A Long Short War’, but I pressed the ‘why did we recommend this’ link. It had been recommended because I’d bought a Warren Zevon album! He did write the only rock song I know of which mentions Saddam Hussein…
Tim Oren 02.09.05 at 7:12 pm
Firefly was acquired by Microsoft, back in 1995. They seem to have buried it.
I wasn’t all that impressed at the time, but part of the reason was the lack of coverage of content I cared about. That was in turn due to a lack of training data from people with tastes similar to mine. All of these taste matching algorithms get ‘smarter’ if they have more diverse data from observing peoples’ behavior. Amazon now has over eight years of observations and contributions from a large customer base with incredibly diverse tastes. That’s called competitive advantage.
Comments on this entry are closed.