Application of Kernels to  Link Analysis  Takahiko Ito  † Masashi Shimbo  † Taku Kudo  ‡ Yuji Matsumoto  † †   Nara Instit...
Background: Link analysis <ul><li>PageRank  and  HITS </li></ul><ul><li>evaluate the  importance  of web pages. </li></ul>...
<ul><li>Kernels defining inner products of nodes in a graph. </li></ul><ul><ul><li>Diffusion kernels  [Chung, 1997 ;  Kond...
Objective <ul><li>To give an interpretation of diffusion kernels in terms of link analysis. </li></ul>
Results <ul><li>Neumann kernels </li></ul><ul><ul><li>subsume both co-citation coupling relatedness and HITS importance. <...
Results <ul><li>Regularized Laplacian kernels / diffusion kernels </li></ul><ul><ul><li>define a new relatedness measure t...
Upcoming SlideShare
Loading in …5
×

KDD 2005

678 views
612 views

Published on

Slides used in the Eleventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.

http://www.sigkdd.org/kdd2005/

Published in: Technology, Business
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
678
On SlideShare
0
From Embeds
0
Number of Embeds
10
Actions
Shares
0
Downloads
0
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide
  • First I will talk about the background of this work. Link analysis is an essential tool for exploring networked data. For example, PageRank and HITS evaluate the importance of web pages. Co-citation coupling used by CiteSeer to estimate relatedness between papers.
  • On the other hand, several kernels for graphs have been proposed recently. In particular, a family of &amp;quot;diffusion&amp;quot; kernels defines an inner product of nodes in a graph, This family includes Heat kernels, Neumann kernels and regularized Laplacian kernels. Here, we have one question. What do these “inner products” represents when viewed as link analysis measures.
  • In this work, We give an interpretation of some diffusion kernels in terms of link analysis. More specifically, we show the interpretation of Neumann kernels and regularized Laplaian kernels. First We show Neumann kernels provide a unified perspective of relatedness and importance. Then We show regularized Laplacian kernels define a new relatedness measure that overcomes some limitations of traditional relatedness. This topic is discussed later.
  • In this work, We give an interpretation of some diffusion kernels in terms of link analysis. More specifically, we show the interpretation of Neumann kernels and regularized Laplaian kernels. First We show Neumann kernels provide a unified perspective of relatedness and importance. Then We show regularized Laplacian kernels define a new relatedness measure that overcomes some limitations of traditional relatedness. This topic is discussed later.
  • In this work, We give an interpretation of some diffusion kernels in terms of link analysis. More specifically, we show the interpretation of Neumann kernels and regularized Laplaian kernels. First We show Neumann kernels provide a unified perspective of relatedness and importance. Then We show regularized Laplacian kernels define a new relatedness measure that overcomes some limitations of traditional relatedness. This topic is discussed later.
  • KDD 2005

    1. 1. Application of Kernels to Link Analysis Takahiko Ito † Masashi Shimbo † Taku Kudo ‡ Yuji Matsumoto † † Nara Institute of Science and Technology ‡ Google
    2. 2. Background: Link analysis <ul><li>PageRank and HITS </li></ul><ul><li>evaluate the importance of web pages. </li></ul><ul><li>Co-citation coupling </li></ul><ul><li>is used by CiteSeer to estimate relatedness </li></ul><ul><li>between papers. </li></ul>
    3. 3. <ul><li>Kernels defining inner products of nodes in a graph. </li></ul><ul><ul><li>Diffusion kernels [Chung, 1997 ; Kondor & Lafferty, 2002] </li></ul></ul><ul><ul><li>Neumann kernels [Kandola et al., 2003] </li></ul></ul><ul><ul><li>Regularized Laplacian kernels [Smola & Kondor, 2003] </li></ul></ul><ul><ul><li>… </li></ul></ul>Background: Kernels for graphs
    4. 4. Objective <ul><li>To give an interpretation of diffusion kernels in terms of link analysis. </li></ul>
    5. 5. Results <ul><li>Neumann kernels </li></ul><ul><ul><li>subsume both co-citation coupling relatedness and HITS importance. </li></ul></ul><ul><ul><li>They also define a spectrum of intermediate measures between the two. </li></ul></ul>Co-citation coupling HITS Neumann kernels
    6. 6. Results <ul><li>Regularized Laplacian kernels / diffusion kernels </li></ul><ul><ul><li>define a new relatedness measure that overcomes some limitations of traditional relatedness (co-citation and bibliographic coupling) </li></ul></ul>

    ×