March 13, 2003

contemplation/clusters

Posted by Arcane Gazebo at March 13, 2003 1:45 PM You know when you're browsing Amazon and it says at the bottom "People who bought this book also bought X?" There's a study of these connections, and I find it very interesting and somewhat disturbing, though maybe not so surprising. (That link is worth clicking on just for the visual impact of the results.)

The reason I find it disturbing is not the high degree of clustering - I think that's to be expected. Rather, it's that there are only two clusters. I tend to think of "liberal" and "conservative" labels as convenient but limited shorthand; there's no a priori reason why economic conservative opinion should correlate with social conservative opinion, for example. From looking at this data, though, it seems as if a clean division can be made. What it suggests is that large numbers of people are getting ideologically trapped; certain sets of ideas have been artifically lumped together by political coalitions, but having grouped in this way people only see one view on every subject. (I actually think this is an overly strong interpretation; I'll say more about that.)

I'm reminded of a column I once read (I think in the National Review) that presented a secular argument against legalizing gay marriage. I think we can all agree that hostility to gay rights in this country is motivated by religion (or by simple bigotry) so it's hard to imagine what a secular argument would look like. In this case it was weak, convoluted, and at times disingenuous. The author seemed like an intelligent person, so he had to know it was a bogus argument, right? Or maybe he was just surrounded by socially conservative material, to the point that "gay marriage should be illegal" becomes a premise rather than a conclusion.

I do think it's easy to read too much into this study, and here's why: the above chart only shows the strongest connections, and not all of them. What the chart looks like is going to be very dependent on where the researcher sets the threshhold for which connections to illustrate. Maybe there are more books in the center, maybe there's more intermixing between the two clusters, but these are not drawn because they're just slightly less common than the connections to What Went Wrong. In other words, the author has a lot of control over how this data appears, and he's going to pick the most striking version.

Where The Night of the Dance fits in to this chart, I have no idea.

Also, this gives me an idea for a game. (I am certainly not the first to think of this, but still...) The obvious name is "Six Degrees of What Went Wrong". The idea is to pick two books, for example Stupid White Men and Slander, and try to connect them on Amazon through as few steps as possible, using the "People who bought x also bought y" link. Those of you who are really really bored, and have a web browser, I encourage to try this and report the results in the comments. Tags:
Comments

book cluster data

Actually, the creator of the diagram[me] does not have much choice on which links are shown. Amazon only makes public the top five 'also bought' books.

If Amazon provided the top 25 'buddy books' for each book then we would not see such clear boundaries. Maybe then everything would be connected to everything else in 1, 2 or 3 steps -- a big hairball. I'm sure some Univ classes require reading books from both extremes. This would show some direct ties between Ann Coulter and Michael Moore -- a thought they each probably find repugnant. ;-) But that pairing is not common when viewing all purchases from those two authors.

Valdis
http://www.orgnet.com

Posted by: Anonymous | March 21, 2003 2:55 AM
Post a comment