Towards Context-Aware Search by Learning A Very Large Variable Length Hidden Markov Model from Search Logs

Can you learn how to re-rank or suggest similar queries or related url from past query sesssion?

The Microsoft paper "Towards Context-Aware Search by Learning A Very Large Variable Length Hidden Markov Model from Search Logs" defines a HMM model with a variable number of hidden states. The learning process is based on a variant of the EM-algorithm. A clustering pre-process step is adopted for reducing the search space of the model. A large scale computation is proposed based on the map-reduce model.

I think that this paper gives a general solution to many different search problem. Anyway, the precision and recall evaluations leave many room for improving the achieved quality of results.
I wonder if similar methodologies are adopted by the recently launched Yahoo's SearchPad.