Nothing Is New

Via Tim Oren, a reference to a search algorithm very similar to contextual networks, called "Spreading Activation Search". Some hunting around leads me to a 1981 dissertation by Scott Preece, which is not available online, and describes exactly the kind of system we've apparently reinvented 22 years later.

I can see two slight differences from our own contextual networks. The first is in the definition of keywords - it's not clear from the link above what method the author used to pick out keywords, but the fact that he gets 3400 words from 1600 documents indicates that the word list is somewhat constrained (a more typical number would be in the 10-20K word range).

The other difference is that spreading activation search also includes author information in the graph. Which is really cool.

Our own contribution seems to be incorporating part-of-speech tagging and noun phrase extraction. As well as Web hyperlinks, which weren't around in 1981.

My own lesson from the Emerging Technology conference is that everything old is new again. I would love to hear more from anyone familiar with this search technique and its history, while I wait for the dissertation to arrive in the mail. In the meantime I've added references to Preece's work in my notes and slides.