If at first you don't succeed, let the search engine try

I’ve been trying to move my research focus in Web searching from descriptive to predictive models. Nothing wrong with descriptive models (we need to understand the world around us) but at some point there is enough descriptive understanding where, in the technology fields, we can develop models that predict or influence.

If at first you don't succeed, let the search engine try is a press release concerning some co-research that I did to develop a predictive model of query reformulation using n-grams. One of my students, Danielle Booth, was a major contributor to this work.

We automatically classified queries within individual sessions in one of 8 states, and then used n-grams to develop Markov chains representing these query reformulation patterns. This search engine log, which was from Dogpile, also recorded when users accessed the query reformulation assistance that most major search engines now offer.

Some finding are:
- the patterns are short (3 – 4 states)
- there is a norm to move to narrow the scope of the query
- there appears to be a cognitive space at the start of the session and when switch vertical where the user is open to system assistance

The research is reported in: Jansen, B.J., Booth, D.L., Spink, A. (2009) Patterns of query reformulation during Web searching. Journal of the American Society for Information Science and Technology. 60(7), 1358-1371.

Thanks to Infospace for access to the logs!