Learning query rewrite policies

A blended query rewrite provider takes as an input a search query and produces a list of query rewrite suggestions. The blended query rewrite provider generates suggestions by applying a number of individual query rewrite providers in series to a query. The sequence in which query rewrite providers are applied to a query is encoded in a query rewrite policy. Policies differ between query types. A policy application engine component selects which policy to apply to a given query based on the cluster to which a query belongs. The policy application engine uses a machine learning algorithm to assign a query to a cluster based on query features. The policy generation engine generates policies and trains the clustering algorithm used by the policy application engine offline based on training data sets and query logs.