Users of mail or news readers are used to sorting content based on source – e.g. from addresses, maybe subject lines, etc. – but this has traditionally been quite clunky and inelegant.

This idea is quite simply to take either Bayesian or LSI filtering currently used to separate spam from legitimate content, and to apply it in a much broader sense.

For example, I might wish to catalogue information about buy-outs or IPOs into one folder, articles about music I might like into another, articles and mails about events into another. By moving articles or mail into folders, the client learns how to automatically filter, much like a junk mail filter learns. That means no matter what the source, information can be categorised non-programatically.

In addition, a reader aware of other sources might be able to identify content from other sources I’m not subscribed to that I am likely to enjoy, because it would score highly in the context of one of my automatic category folders.