Featurerank: A non-linear listwise approach with clustering and boosting
Yongqing Wang, Wenji Mao · 2010
Listwise is an important approach in learning to rank. Most of the existing lisewise methods use a linear ranking function which can only achieve a limited performance being applied to complex ranking problem. This paper proposes a non-linear listwise algorithm inspired by boosting and clustering. Different from the previous listwise approaches, our algorithm constructs weak rankers through directly discovering hidden order in single feature, and then combines these weak rankers using a boosting procedure. To discover the hidden order, we utilize (KNN) method. In our preliminary experiment, we compare our approach with other listwise algorithms and show the effectiveness of our proposed algorithm.