Improved performance of the identification.

This is done by splitting the scoring part of the query and using a
filter for the type. Experiments proved a substantial improvement.