What they've ended up producing is a Java application that takes an initial search term (in books, music or DVDs), generates a similarities map, and then makes it clickable. Click once on the letter 'H' to get a description of the item, click twice on the item bubble to generate similarities for this item. It's all done in a clean, fast, well designed UI. You can scale the display, and drag results around the screen in a dynamic, elastic environment, which gives this completely abstract application a certain physicality.
This is is exactly the sort of thing that web services should be used for at this stage of their development - inspirational tools that show a whole new way of working with existing data sets.
Well worth an exploration. Or three. You can wander around whole geographies of ideas, without leaving your browser.