Application of kernels to link analysis

Abstract. The application of kernel methods to link analysis is explored. We argue that a family of kernels on graphs provides a unified perspective on the three measures proposed for link analysis: relatedness, global importance, and relative importance. The framework provided by the kernels establishes relative importance as an intermediate between relatedness and global importance, in which the bias between relatedness and importance is naturally controlled by a parameter characterizing individual kernels. An experimental evaluation of the property of these measures is carried out using a real-world network of bibliographic citations.

• Type: Conference
• Year: 2005
• Where: KDD
• Authors: Takahiko Ito, Masashi Shimbo, Taku Kudo, Yuji Matsumoto
• Keywords: Data Mining, Global Importance, Individual Kernels, KDD 2005, Relative Importance