Visualizing Last.fm's Friends Network
Last.fm, one of the web's most popular recommendation and web radio services, provides their recommendations based on what "people like you" enjoy. This brings a social aspect to their system where friendships have an impact what people listen to. But what do these friendships look like? Are they localized groups having little contact with other users? Or do they actually span across the network of last.fm users?
An academic blogger, Anonymous Prof, wanted to find out.
Using the Tulip visualization package, he sampled 25,000 relationships out of a data set of 166,332 users pulled from last.fm with help from their API. From this data, he came up with 2310 seed users with 19,008 friends, resulting in 24,036 relationships and an average of 10.41 friends.
From the resulting visual graph, it was clear that last.fm's user network is actually very strong. Although there are some clusters of close friends, even those people have a lot of interconnectivity:
Along the outside of the main network, he found users who either did not have connections within the main network or whose connections were missing because he was only looking at a sampling of users as opposed to the full network. Likely, it was a combination of both. Interestingly enough, even among these outlying users, they had smaller networks of their own:
Graphing the distribution of users to number of friends, it was clear that the majority of last.fm users only have a few friends:
All this data was released on the Anonymous Prof's blog yesterday, where you can find even more images and details. Next up, he plans on collecting the listening history of users in order to examine music listening patterns as they relate to the network. Visualization junkies will want to stay tuned for that.
Share
Comments
Subscribe to comments for this post OR Subscribe to comments for all ReadWriteWeb posts
Nice graph :)
How long did it take to go through the whole data?
By the way, you have a little spelling error: anoNymous - not anoymous ;)
yep, math applies to everything.
i'd say it was about time some math to publicly enter the social network cosmos, as networks/matrices of objects are the favorite subject of higher mathematics.
just look what math did to the games sector, and you can start imagining the future of social networks analysis
I love Last.fm for this reason. It's a great model of how recommendations should be made in my opinion.
@GRiNSER: corrected - thank you!
Interesting data analysis by Anonymous Prof. Last.fm is one of the few websites that has a lot of functionality yet their user interface still seems to be clean and refreshing. More importantly, it's a 'social network' that is actually USEFUL. They are ahead of the game, and always have been.
Hi Sarah,
Thanks for the link!
To answer GRiNSER's question. The data collection is still on going and should take a total of about 2 weeks. As for processing time, converting the raw data from Last.fm into something usable took about 5 minutes and then Tulip did the graphing in just a minute or two...nothing too crazy.
I've also gotten a great suggestion from a reader to use a different software package (GUESS) that should handle more data. If i can get it up and running this weekend I'll post visualizations of a larger sample.
-AnonymousProf
It'd be interesting to see this work expanded to other niche and even broader social networks. I'd love to see relationship and user connection comparisons across the realm of social networks.
I'd say the reason the networks are small is that Last.fm is still new in the mainstream, and users are more concerned initially with just finding music first and foremost. Expect these networks to grow as more users see the perks in finding friends with similar tastes.