Toward Representation Independent Similarity Search Over Graphs - - PowerPoint PPT Presentation

toward representation independent
SMART_READER_LITE
LIVE PREVIEW

Toward Representation Independent Similarity Search Over Graphs - - PowerPoint PPT Presentation

Toward Representation Independent Similarity Search Over Graphs Yodsawalai Chodpathumwan , University of Illinois at Urbana-Champaign Arash Termehchy , Oregon State University Yizhou Sun , Northeastern University Amirhossein Aleyasin , University


slide-1
SLIDE 1

Toward Representation Independent Similarity Search Over Graphs

Yodsawalai Chodpathumwan, University of Illinois at Urbana-Champaign Arash Termehchy, Oregon State University Yizhou Sun, Northeastern University Amirhossein Aleyasin, University of Illinois at Urbana-Champaign Jose Picado, Oregon State University

slide-2
SLIDE 2

Similarity Search over Graph Databases

Which movies are similar to “Casino Royale” in IMDb?

1

Lord Asriel

character

Daniel Craig

actor

James Bond

character

Martin Campbell

director

Casino Royale

movie

Golden Compass

movie

Green Lantern

movie

Mask of Zorro

movie

Answers Golden Compass Green Lantern Mask of Zorro Query Casino Royale

slide-3
SLIDE 3

Same information is represented in many ways

2

Lord Asriel

character

Daniel Craig

actor

James Bond

character

Martin Campbell

director

Casino Royale

movie

Golden Compass

movie

Green Lantern

movie

Mask of Zorro

movie directed-by

Casino Royale

movie

Martin Campbell

director

James Bond

character

Daniel Craig

actor

Lord Asriel

character

Golden Compass

movie

Mask of Zorro

movie

Green Lantern

movie starring starring directed by directed by

slide-4
SLIDE 4

3

  • J. Han

author

  • P. S. Yu

author

Truth Discovery

paper

RankClus

paper

KDD

conference

EDBT

conference

2009

year

2007

year

  • P. Zhao, CIKM’09
  • Y. Sun, PVLDB’11

Same information is represented in many ways

  • J. Han

author

  • P. S. Yu

author

Truth Discovery

paper

RankClus

paper

KDD

conference

EDBT

conference

2009

year

2007

year

slide-5
SLIDE 5

Which movies are similar to “Casino Royale” ?

4

SimRank [Jeh, KDD’02] Golden Compass Mask of Zorro, Green Lantern Lord Asriel

character

Daniel Craig

actor

James Bond

character

Martin Campbell

director

Casino Royale

movie

Golden Compass

movie

Green Lantern

movie

Mask of Zorro

movie

slide-6
SLIDE 6

Which movies are similar to “Casino Royale” ?

5

SimRank Mask of Zorro, Green Lantern Golden Compass Casino Royale

movie

Martin Campbell

director

James Bond

character

Daniel Craig

actor

Lord Asriel

character

Golden Compass

movie

Mask of Zorro

movie

Green Lantern

movie

starring starring directed-by directed-by directed-by

slide-7
SLIDE 7

Generality of Similarity Search over Graphs

  • 6
slide-8
SLIDE 8

Invertible Transformation

  • 7

Daniel Craig

actor

James Bond

character

Casino Royale

movie

Casino Royale

film

James Bond

character

Daniel Craig

actor

starring

T T –1

slide-9
SLIDE 9

Information Preserving Transformation

  • 8

Daniel Craig

actor

James Bond

character

Casino Royale

movie

Casino Royale

film

James Bond

character

Daniel Craig

actor

starring

T T –1

slide-10
SLIDE 10

Information Equivalent Graphs

Databases D1 and D2 are information equivalent iff there exists an information preserving transformation from D1 to D2 and vice versa.

9

Daniel Craig

actor

James Bond

character

Casino Royale

movie

Casino Royale

film

James Bond

character

Daniel Craig

actor

starring

slide-11
SLIDE 11

10

Lord Asriel

character

Daniel Craig

actor

James Bond

character

Martin Campbell

director

Casino Royale

movie

Golden Compass

movie

Green Lantern

movie

Mask of Zorro

movie

SimRank Golden Compass Mask of Zorro, Green Lantern SimRank Mask of Zorro, Green Lantern Golden Compass

SimRank is not general

directed-by

Casino Royale

movie

Martin Campbell

director

James Bond

character

Daniel Craig

actor

Lord Asriel

character

Golden Compass

movie

Mask of Zorro

movie

Green Lantern

movie starring starring directed by directed by

slide-12
SLIDE 12

Evaluation on Generality of Similarity Search Algorithms

11

DB

(Movies,Bib)

Q

30 queries/DB

T(DB)

(Movies,Bib)

Algorithms

SimRank, RWR, PathSim

Ranked answers Ranked answers

T

Kendall’s Tau

slide-13
SLIDE 13

Average Ranking Differences

12

  • Top 10 answers
  • Top 50 answers

Algorithms Movie DBs Bib DBs RWR [Tong, KDD’06] 0.375 0.773 SimRank [Jeh, KDD’02] 0.418 0.626 PathSim [Sun, VLDB’11] 0.375 0.953 Algorithms Movie DBs Bib DBs RWR [Tong, KDD’06] 0.204 0.718 SimRank [Jeh, KDD’02] 0.264 0.673 PathSim [Sun, VLDB’11] 0.110 0.688

slide-14
SLIDE 14

Conclusion

  • Introduced the problem of representation independent similarity

search over graph data.

  • Proposed a formal framework to measure generality
  • Performed empirical study and showed that some of the well-known

similarity search algorithms are not general.

13

slide-15
SLIDE 15
slide-16
SLIDE 16

Unused Slides

slide-17
SLIDE 17

When PathSim fails?

slide-18
SLIDE 18

Movie Databases Bibliography Databases

Transformations

17

IMDb [P. Zhao, CIKM’09] Freebase [Y. Sun, PVLDB’11]