Effect on generalization of using relational information in list-wise algorithms

Abstract

Learning to rank became a hot research topic in recent years and utilizing relational information in list-wise algorithms was discovered to be valuable and was widely adopted in various algorithms. These algorithms' empirical performances were usually given, but few of them conduct theoretical analysis on the generalization bound. Based on the theory of Rademacher Average, we derive the generalization bound of ranking relational objects algorithms and discuss the effect on the generalization bound of using this method. Especially, an interesting property of ranking relational objects algorithms for Topic Distillation was discovered: the generalization bound does not depend on the size of documents in each query in training set. Experiments are conducted to verify this property. © 2013 Springer-Verlag.