|
Abstract:
|
In many areas of science, data are given in graphs, Where vertices correspond to objects and edges encode similarity among objects. Ranking graph data has been an important issue both theoretically and practically, driven by the demand of fast and efficient information processing such as the search engine in computer technology. In this talk, we propose a robust ranking method for graph data based on the recently introduced kernelized spatial depth function (Chen, Dang, Peng, Butler (2007)). KSD generalizes the spatial depth function via a Hilbert space reproducing kernel. We compare the proposed KSD-based ranking with other existing rankings. Ours demonstrates competitive performance mainly because the KSD successfully captures global and local structure of a graph.
|
