Ball Ranking Machine for Content-Based Multimedia Retrieval
Dijun Luo and Heng Huang
This paper presents Ball Ranking Machines (BRMs) to address the supervised ranking problems. In previous work by other researchers, supervised ranking methods have been successfully applied in information retrieval tasks. Among these methodologies, Ranking Support Vector Machines (Rank SVMs) are well studied. However, one major fact which limits the application is that Ranking SVMs need to optimize a margin-based objection function over all possible document pairs within any queries. In consequence, Rank SVMs need to pick a large number of support vectors among a huge number of support vector candidates. This paper modifies the objective function of Ranking SVMs and offers a theoretical analysis which indicates that such objective function must generate a sparse solution according to the KKT conditions. A fast approximation of raking machine algorithm is proposed, which decreases the training time and generates much fewer support vectors. Empirical studies on both synthetic data, content-based image/video retrieval data show that our method is comparable with Ranking SVM, meanwhile using much fewer ranking support vectors and significantly less training time.