A comparative study of methods for estimating query language models with pseudo feedback