Someone sent me a message on Last.fm pointing out that I’m the top fan for A Stroke of Genius by the Freelance Hellraiser (and in fact the top fan for Freelance Hellraiser on the whole of last.fm which surprised me). The other top fan is a user called russeljsmith. One of russeljsmith’s last.fm friends is behemothpuss who is also one of my last.fm friends.

Visiting russeljsmith’s Flickr profile page I see that not only is one of his contacts Leeds Guy, who in turn has me as one of his contacts but that russeljsmith and behemothpuss (Emma B on Flickr) are mutual Flickr contacts, and Emma B and I are mutual contacts.

It’s quite possible that there are a number of other points of contact between myself and this person which I’ll never discover.

Looking at the categories on his weblog, we have quite a few interests in common (web development, usability, games), although he doesn’t post frequently enough for me to actually subscribe. What’s really intriguing is the prospect that there may be other people I’m linked to in a similar manner, but will never find.

There’s virtually no chance that without the initial prompting I would ever have discovered russeljsmith, so how could I have done so automatically?

Flickr-FOAF tools could have revealed the links, but they’d also reveal links to many, many other people as well. I don’t think there are currently any tools to extract a list of people with common musical taste from last.fm, or rather, other top fans of bands for whom you are a massive fan (it does have a set of webservices so these could possibly be leveraged to provide what’s necessary, but I’m not sure – just getting a list of overall “musical neighbours” for a user isn’t enough). After that, if you still have a list of, say, thirty people then the only thing for it is to run their homepages through a keyword extraction tool like Yahoo’s (there’s another popular one that launched recently but I can’t quite remember what it is – it seemed to judge my blog as being about Web 2.0 which, as you can imagine, made me terribly bitter 😉 ) and compare the keywords against a list of your interests. Still, I can’t imagine that the results are likely to be any better than shaky.

What else could be done to mine these systems for user-user recommendations?