Title :
Stability analysis for ranking algorithms
Author :
Gao, Wei ; Zhang, Yungang ; Liang, Li ; Xia, Youming
Author_Institution :
Dept. of Inf., Yunnan Normal Univ., Kunming, China
Abstract :
In this paper, the stability of ranking algorithms is studied by adopting a strategy which adjusts the sample set by deleting one or two element from it. Relationship between uniform loss stability and uniform score stability is investigated. A sufficient condition for uniform score stability is given. The result of our work shows that if a uniform score stability ranking algorithm use γ ranking loss, then it has uniform loss stability; also, if for any x, a kernel function K(x, x) has a limited upper bound, then the ranking algorithm which minimizes the regularization empirical l-error will have good uniform score stability.
Keywords :
algorithm theory; information retrieval; search engines; ranking algorithm stability; uniform loss stability; uniform score stability; Algorithm design and analysis; Kernel; Machine learning algorithms; Manganese; Stability analysis; Thermal stability; Training; RKHS; ranking; ranking loss function; uniform loss stability; uniform score stability;
Conference_Titel :
Information Theory and Information Security (ICITIS), 2010 IEEE International Conference on
Conference_Location :
Beijing
Print_ISBN :
978-1-4244-6942-0
DOI :
10.1109/ICITIS.2010.5689665
