The basic idea of AdaRank is constructing “weak rankers” repeatedly based on reweighted training queries and linearly combining the weak rankers for making ranking predictions. In learning, AdaRank minimizes a loss function directly defined on performance measures. The details of AdaRank can be found in the paper “AdaRank: A Boosting Algorithm for Information Retrieval.”