KernelRank: Exploiting semantic linkage kernels for relevant pages finding

Relevant pages finding is to find a set of relevant pages that address the same topic as the given page. Hyperlink relationship is an important useful clue for this task. Some hyperlinks are useful, also some are irrelevant or noisy. Therefore, it is important to design efficient relevant pages finding methods that can work well in the real-world Web data. In this paper, we propose a relevant pages finding algorithm, KernelRank. This algorithm takes advantage of linkage kernels to reveal latent semantic relationships among pages and to identify relevant pages precisely and effectively. Experiments are conducted on WT10G and the results show that the KernelRank algorithm is feasible and effective.