Combining Learning-to-Rank with Clustering

This paper aims to combine learning-to-rank methods with an existing clustering underlying the entities to be ranked. In recent years, learning-to-rank has attracted the interest of many researchers and a large number of algorithmic approaches and methods have been published. Existing learning-to-rank methods have as goal to automatically construct a ranking model from training data. Usually, all these methods don't take into consideration the data's structure. Although there is a novel task named “Relational Ranking” which tries to make allowances for the inter-relationship between documents, it has restrictions and it is difficult to be applied in a lot of real applications. To address this problem, we create a per query clustering using state of the art algorithms from our training data. Then, we experimentally verify the effect of