One of the most important abilities of human is cluster and classify similar things, which makes people could better understand the nature, easier establish and manage the social society. How to model things like people and how to compute the similarities between models are two major problems need to be solved to make the machine has this ability. For the first problem, the Semantic Link Network (SLN) could be a appropriate choice, however, the challenge is how to compute the similarities between SLNs. In this paper, we design a hierarchical graph kernel for SLNs to solve this challenge. We evaluate the practical performance of our kernels on a task of detecting semantic similar relationships between texts. The result show that the detection results of semantic node hierarchical kernel is most similar to human's.