SlideShare une entreprise Scribd logo
1  sur  14
Télécharger pour lire hors ligne
Web Science & Technologies
                          University of Koblenz ▪ Landau, Germany




  Predicting Directed Links using
Nondiagonal Matrix Decompositions
           Jérôme Kunegis & Jörg Fliege


           Int. Conf. on Data Mining 2012
Trust Prediction




                                                        ?




                    Goal: predict trusted edges

           Jérôme Kunegis & Jörg Fliege   Predicting Directed Links using Nondiagonal Matrix Decompositions
           kunegis@uni-koblenz.de         ICDM 2012                                               2
Triangle Closing




                                                ?




           Jérôme Kunegis & Jörg Fliege   Predicting Directed Links using Nondiagonal Matrix Decompositions
           kunegis@uni-koblenz.de         ICDM 2012                                               3
Powers of the Adjacency Matrix

         1                     2


 3                                          4


         5                 6                                          0     1     1     0     1     1
                                                                      0     0     0     1     0     0
                                                                      0     0     0     0     1     1
     (A²)14 = 2                                      A=               0     0     1     0     0     1
                                                                      1     0     0     0     0     0
     (A³)14 = 1                                                       0     0     0     1     0     0

             Jérôme Kunegis & Jörg Fliege       Predicting Directed Links using Nondiagonal Matrix Decompositions
             kunegis@uni-koblenz.de             ICDM 2012                                               4
Computing Ak When A is Symmetric


 Eigenvalue decomposition:                                                  A=UΛU                        T




 Ak = (U Λ UT) (U Λ UT) . . . (U Λ UT)
                                         k   T
                      =UΛ U


                                    Problem: A is asymmetric


          Jérôme Kunegis & Jörg Fliege           Predicting Directed Links using Nondiagonal Matrix Decompositions
          kunegis@uni-koblenz.de                 ICDM 2012                                               5
Asymmetric Eigenvalue Decomposition


When A is diagonalizable:                                          A=UΛU                        −1




  Problem:
   A is not
diagonalizable                                                            Advogato




          Jérôme Kunegis & Jörg Fliege   Predicting Directed Links using Nondiagonal Matrix Decompositions
          kunegis@uni-koblenz.de         ICDM 2012                                               6
Singular Value Decomposition

                                                                                                         T
                                                                             A=UΛV

                                 k            T
                      UΣ V
                 T                        T                             T
   =UΣV VΣU ...UΣV
                                 T
               =AA ...A
                                                                    Problem: This does not
                                                                           equal Ak


           Jérôme Kunegis & Jörg Fliege           Predicting Directed Links using Nondiagonal Matrix Decompositions
           kunegis@uni-koblenz.de                 ICDM 2012                                               7
DEDICOM


                                                                                            T
 Solution:                                               A=UXU


 “DEDICOM – Decomposition into Directed Components”



   X=                                                             Not diagonal

                 Advogato
          Jérôme Kunegis & Jörg Fliege   Predicting Directed Links using Nondiagonal Matrix Decompositions
          kunegis@uni-koblenz.de         ICDM 2012                                               8
Computation of Ak with DEDICOM




      A =UX Uk                                                k                       T


                               k
                         A is easy to compute



          Jérôme Kunegis & Jörg Fliege   Predicting Directed Links using Nondiagonal Matrix Decompositions
          kunegis@uni-koblenz.de         ICDM 2012                                               9
Finding a DEDICOM



   Singular value decomposition:



                                                  T
                    A = U(Σ V U ) U                                  T




                                     Problem: Not computed to full rank


          Jérôme Kunegis & Jörg Fliege    Predicting Directed Links using Nondiagonal Matrix Decompositions
          kunegis@uni-koblenz.de          ICDM 2012                                               10
DEDICOM Algorithms

                                Using singular value decomposition A = U Σ VT



LEFT        A = U (Σ VT U) UT
RIGHT       A = V (VT U Σ) VT
CLO         A = Q X QT
                U Σ UT + V Σ VT = Q Λ QT (eigenvalue decomp.)
                X = QT A Q
ITER        Iterative algorithm



(Harshman 1978)                          (Kliers et al. 1990)
         Jérôme Kunegis & Jörg Fliege     Predicting Directed Links using Nondiagonal Matrix Decompositions
         kunegis@uni-koblenz.de           ICDM 2012                                               11
Approximation of eA = I + A + ½A² + ⅙A³ + . . .

         Approximating A                                               Approximating eA




      Advogato trust
                                                                   Advogato trust




                                                                DEDICOM algorithms


                  Jérôme Kunegis & Jörg Fliege   Predicting Directed Links using Nondiagonal Matrix Decompositions
                  kunegis@uni-koblenz.de         ICDM 2012                                               12
Jérôme Kunegis
kunegis@uni-koblenz.de

Jörg Fliege
j.fliege@soton.ac.uk




                         Thank You

 konect.uni-koblenz.de
References



    Predicting directed links using nondiagonal matrix decompositions
    Jérôme Kunegis & Jörg Fliege
    Int. Conf. on Data Mining, 2012

    Models for analysis of asymmetrical relationships among n objects or stimuli
    Richard A. Harshman
    Contributions to economic analysis 187, 185–204, 1990

    A generalization of Takane's algorithm for DEDICOM
    Henk A. Kliers, Jos M. ten Berge, Yoshio Takane & Jan de Leeuw
    Psychometrika 55(1), 151–158, 1990




              Jérôme Kunegis & Jörg Fliege   Predicting Directed Links using Nondiagonal Matrix Decompositions
              kunegis@uni-koblenz.de         ICDM 2012                                               14

Contenu connexe

Plus de Jérôme KUNEGIS

Generating Networks with Arbitrary Properties
Generating Networks with Arbitrary PropertiesGenerating Networks with Arbitrary Properties
Generating Networks with Arbitrary Properties
Jérôme KUNEGIS
 
Preferential Attachment in Online Networks: Measurement and Explanations
Preferential Attachment in Online Networks:  Measurement and ExplanationsPreferential Attachment in Online Networks:  Measurement and Explanations
Preferential Attachment in Online Networks: Measurement and Explanations
Jérôme KUNEGIS
 
Why Beyoncé Is More Popular Than Me – Fairness, Diversity and Other Measures
Why Beyoncé Is More Popular Than Me – Fairness, Diversity and Other MeasuresWhy Beyoncé Is More Popular Than Me – Fairness, Diversity and Other Measures
Why Beyoncé Is More Popular Than Me – Fairness, Diversity and Other Measures
Jérôme KUNEGIS
 
Fairness on the Web: Alternatives to the Power Law (WebSci 2012)
Fairness on the Web:  Alternatives to the Power Law (WebSci 2012)Fairness on the Web:  Alternatives to the Power Law (WebSci 2012)
Fairness on the Web: Alternatives to the Power Law (WebSci 2012)
Jérôme KUNEGIS
 
Searching Microblogs: Coping with Sparsity and Document Quality
Searching Microblogs: Coping with Sparsity and Document QualitySearching Microblogs: Coping with Sparsity and Document Quality
Searching Microblogs: Coping with Sparsity and Document Quality
Jérôme KUNEGIS
 

Plus de Jérôme KUNEGIS (19)

Generating Networks with Arbitrary Properties
Generating Networks with Arbitrary PropertiesGenerating Networks with Arbitrary Properties
Generating Networks with Arbitrary Properties
 
Karriere Lounge – INFORMATIK 2013
Karriere Lounge – INFORMATIK 2013Karriere Lounge – INFORMATIK 2013
Karriere Lounge – INFORMATIK 2013
 
Eight Formalisms for Defining Graph Models
Eight Formalisms for Defining Graph ModelsEight Formalisms for Defining Graph Models
Eight Formalisms for Defining Graph Models
 
What Is the Added Value of Negative Links in Online Social Networks?
What Is the Added Value of Negative Links in Online Social Networks?What Is the Added Value of Negative Links in Online Social Networks?
What Is the Added Value of Negative Links in Online Social Networks?
 
KONECT – The Koblenz Network Collection
KONECT – The Koblenz Network CollectionKONECT – The Koblenz Network Collection
KONECT – The Koblenz Network Collection
 
Preferential Attachment in Online Networks: Measurement and Explanations
Preferential Attachment in Online Networks:  Measurement and ExplanationsPreferential Attachment in Online Networks:  Measurement and Explanations
Preferential Attachment in Online Networks: Measurement and Explanations
 
Online Dating Recommender Systems: The Split-complex Number Approach
Online Dating Recommender Systems: The Split-complex Number ApproachOnline Dating Recommender Systems: The Split-complex Number Approach
Online Dating Recommender Systems: The Split-complex Number Approach
 
Why Beyoncé Is More Popular Than Me – Fairness, Diversity and Other Measures
Why Beyoncé Is More Popular Than Me – Fairness, Diversity and Other MeasuresWhy Beyoncé Is More Popular Than Me – Fairness, Diversity and Other Measures
Why Beyoncé Is More Popular Than Me – Fairness, Diversity and Other Measures
 
Fairness on the Web: Alternatives to the Power Law (WebSci 2012)
Fairness on the Web:  Alternatives to the Power Law (WebSci 2012)Fairness on the Web:  Alternatives to the Power Law (WebSci 2012)
Fairness on the Web: Alternatives to the Power Law (WebSci 2012)
 
Fairness on the Web: Alternatives to the Power Law
Fairness on the Web:  Alternatives to the Power LawFairness on the Web:  Alternatives to the Power Law
Fairness on the Web: Alternatives to the Power Law
 
KONECT Cloud – Large Scale Network Mining in the Cloud
KONECT Cloud – Large Scale Network Mining in the CloudKONECT Cloud – Large Scale Network Mining in the Cloud
KONECT Cloud – Large Scale Network Mining in the Cloud
 
On the Spectral Evolution of Large Networks (PhD Thesis by Jérôme Kunegis)
On the Spectral Evolution of Large Networks (PhD Thesis by Jérôme Kunegis)On the Spectral Evolution of Large Networks (PhD Thesis by Jérôme Kunegis)
On the Spectral Evolution of Large Networks (PhD Thesis by Jérôme Kunegis)
 
Searching Microblogs: Coping with Sparsity and Document Quality
Searching Microblogs: Coping with Sparsity and Document QualitySearching Microblogs: Coping with Sparsity and Document Quality
Searching Microblogs: Coping with Sparsity and Document Quality
 
Bad News Travel Fast: A Content-based Analysis of Interestingness on Twitter
Bad News Travel Fast: A Content-based Analysis of Interestingness on TwitterBad News Travel Fast: A Content-based Analysis of Interestingness on Twitter
Bad News Travel Fast: A Content-based Analysis of Interestingness on Twitter
 
On the Scalability of Graph Kernels Applied to Collaborative Recommenders
On the Scalability of Graph Kernels Applied to Collaborative RecommendersOn the Scalability of Graph Kernels Applied to Collaborative Recommenders
On the Scalability of Graph Kernels Applied to Collaborative Recommenders
 
Learning Spectral Graph Transformations for Link Prediction
Learning Spectral Graph Transformations for Link PredictionLearning Spectral Graph Transformations for Link Prediction
Learning Spectral Graph Transformations for Link Prediction
 
The Slashdot Zoo: Mining a Social Network with Negative Edges
The Slashdot Zoo:  Mining a Social Network with Negative EdgesThe Slashdot Zoo:  Mining a Social Network with Negative Edges
The Slashdot Zoo: Mining a Social Network with Negative Edges
 
Spectral Analysis of Signed Graphs for Clustering, Prediction and Visualization
Spectral Analysis of Signed Graphs for Clustering, Prediction and VisualizationSpectral Analysis of Signed Graphs for Clustering, Prediction and Visualization
Spectral Analysis of Signed Graphs for Clustering, Prediction and Visualization
 
Network Growth and the Spectral Evolution Model
Network Growth and the Spectral Evolution ModelNetwork Growth and the Spectral Evolution Model
Network Growth and the Spectral Evolution Model
 

Dernier

Top Rated Call Girls In Podanur 📱 {7001035870} VIP Escorts Podanur
Top Rated Call Girls In Podanur 📱 {7001035870} VIP Escorts PodanurTop Rated Call Girls In Podanur 📱 {7001035870} VIP Escorts Podanur
Top Rated Call Girls In Podanur 📱 {7001035870} VIP Escorts Podanur
dharasingh5698
 
VIP Call Girls Mehsana 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Mehsana 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Mehsana 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Mehsana 7001035870 Whatsapp Number, 24/07 Booking
dharasingh5698
 
Corporate Presentation Probe Canaccord Conference 2024.pdf
Corporate Presentation Probe Canaccord Conference 2024.pdfCorporate Presentation Probe Canaccord Conference 2024.pdf
Corporate Presentation Probe Canaccord Conference 2024.pdf
Probe Gold
 
B2 Interpret the brief.docxccccccccccccccc
B2 Interpret the brief.docxcccccccccccccccB2 Interpret the brief.docxccccccccccccccc
B2 Interpret the brief.docxccccccccccccccc
MollyBrown86
 
Call Girls Chandigarh Just Call 8868886958 Top Class Call Girl Service Available
Call Girls Chandigarh Just Call 8868886958 Top Class Call Girl Service AvailableCall Girls Chandigarh Just Call 8868886958 Top Class Call Girl Service Available
Call Girls Chandigarh Just Call 8868886958 Top Class Call Girl Service Available
Sheetaleventcompany
 
Corporate Presentation Probe May 2024.pdf
Corporate Presentation Probe May 2024.pdfCorporate Presentation Probe May 2024.pdf
Corporate Presentation Probe May 2024.pdf
Probe Gold
 

Dernier (20)

Top Rated Call Girls In Podanur 📱 {7001035870} VIP Escorts Podanur
Top Rated Call Girls In Podanur 📱 {7001035870} VIP Escorts PodanurTop Rated Call Girls In Podanur 📱 {7001035870} VIP Escorts Podanur
Top Rated Call Girls In Podanur 📱 {7001035870} VIP Escorts Podanur
 
VIP Call Girls Mehsana 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Mehsana 7001035870 Whatsapp Number, 24/07 BookingVIP Call Girls Mehsana 7001035870 Whatsapp Number, 24/07 Booking
VIP Call Girls Mehsana 7001035870 Whatsapp Number, 24/07 Booking
 
Corporate Presentation Probe Canaccord Conference 2024.pdf
Corporate Presentation Probe Canaccord Conference 2024.pdfCorporate Presentation Probe Canaccord Conference 2024.pdf
Corporate Presentation Probe Canaccord Conference 2024.pdf
 
countries with the highest gold reserves in 2024
countries with the highest gold reserves in 2024countries with the highest gold reserves in 2024
countries with the highest gold reserves in 2024
 
Dattawadi ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready Fo...
Dattawadi ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready Fo...Dattawadi ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready Fo...
Dattawadi ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready Fo...
 
Teekay Tankers Q1-24 Earnings Presentation
Teekay Tankers Q1-24 Earnings PresentationTeekay Tankers Q1-24 Earnings Presentation
Teekay Tankers Q1-24 Earnings Presentation
 
Collective Mining | Corporate Presentation - May 2024
Collective Mining | Corporate Presentation - May 2024Collective Mining | Corporate Presentation - May 2024
Collective Mining | Corporate Presentation - May 2024
 
ITAU EQUITY_STRATEGY_WARM_UP_20240505 DHG.pdf
ITAU EQUITY_STRATEGY_WARM_UP_20240505 DHG.pdfITAU EQUITY_STRATEGY_WARM_UP_20240505 DHG.pdf
ITAU EQUITY_STRATEGY_WARM_UP_20240505 DHG.pdf
 
Call Girls Marunji Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Marunji Call Me 7737669865 Budget Friendly No Advance BookingCall Girls Marunji Call Me 7737669865 Budget Friendly No Advance Booking
Call Girls Marunji Call Me 7737669865 Budget Friendly No Advance Booking
 
Western Copper and Gold - May 2024 Presentation
Western Copper and Gold - May 2024 PresentationWestern Copper and Gold - May 2024 Presentation
Western Copper and Gold - May 2024 Presentation
 
Teck Supplemental Information, May 2, 2024
Teck Supplemental Information, May 2, 2024Teck Supplemental Information, May 2, 2024
Teck Supplemental Information, May 2, 2024
 
Balaji Nagar ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready...
Balaji Nagar ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready...Balaji Nagar ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready...
Balaji Nagar ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready...
 
Teekay Corporation Q1-24 Earnings Results
Teekay Corporation Q1-24 Earnings ResultsTeekay Corporation Q1-24 Earnings Results
Teekay Corporation Q1-24 Earnings Results
 
B2 Interpret the brief.docxccccccccccccccc
B2 Interpret the brief.docxcccccccccccccccB2 Interpret the brief.docxccccccccccccccc
B2 Interpret the brief.docxccccccccccccccc
 
SME IPO Opportunity and Trends of May 2024
SME IPO Opportunity and Trends of May 2024SME IPO Opportunity and Trends of May 2024
SME IPO Opportunity and Trends of May 2024
 
Call Girls Chandigarh Just Call 8868886958 Top Class Call Girl Service Available
Call Girls Chandigarh Just Call 8868886958 Top Class Call Girl Service AvailableCall Girls Chandigarh Just Call 8868886958 Top Class Call Girl Service Available
Call Girls Chandigarh Just Call 8868886958 Top Class Call Girl Service Available
 
High Profile Call Girls in Pune (Adult Only) 8005736733 Escort Service 24x7 ...
High Profile Call Girls in Pune  (Adult Only) 8005736733 Escort Service 24x7 ...High Profile Call Girls in Pune  (Adult Only) 8005736733 Escort Service 24x7 ...
High Profile Call Girls in Pune (Adult Only) 8005736733 Escort Service 24x7 ...
 
Vijayawada ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready F...
Vijayawada ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready F...Vijayawada ( Call Girls ) Pune  6297143586  Hot Model With Sexy Bhabi Ready F...
Vijayawada ( Call Girls ) Pune 6297143586 Hot Model With Sexy Bhabi Ready F...
 
VVIP Pune Call Girls Parvati Gaon WhatSapp Number 8005736733 With Elite Staff...
VVIP Pune Call Girls Parvati Gaon WhatSapp Number 8005736733 With Elite Staff...VVIP Pune Call Girls Parvati Gaon WhatSapp Number 8005736733 With Elite Staff...
VVIP Pune Call Girls Parvati Gaon WhatSapp Number 8005736733 With Elite Staff...
 
Corporate Presentation Probe May 2024.pdf
Corporate Presentation Probe May 2024.pdfCorporate Presentation Probe May 2024.pdf
Corporate Presentation Probe May 2024.pdf
 

Predicting Directed Links using Nondiagonal Matrix Decompositions

  • 1. Web Science & Technologies University of Koblenz ▪ Landau, Germany Predicting Directed Links using Nondiagonal Matrix Decompositions Jérôme Kunegis & Jörg Fliege Int. Conf. on Data Mining 2012
  • 2. Trust Prediction ? Goal: predict trusted edges Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 2
  • 3. Triangle Closing ? Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 3
  • 4. Powers of the Adjacency Matrix 1 2 3 4 5 6 0 1 1 0 1 1 0 0 0 1 0 0 0 0 0 0 1 1 (A²)14 = 2 A= 0 0 1 0 0 1 1 0 0 0 0 0 (A³)14 = 1 0 0 0 1 0 0 Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 4
  • 5. Computing Ak When A is Symmetric Eigenvalue decomposition: A=UΛU T Ak = (U Λ UT) (U Λ UT) . . . (U Λ UT) k T =UΛ U Problem: A is asymmetric Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 5
  • 6. Asymmetric Eigenvalue Decomposition When A is diagonalizable: A=UΛU −1 Problem: A is not diagonalizable Advogato Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 6
  • 7. Singular Value Decomposition T A=UΛV k T UΣ V T T T =UΣV VΣU ...UΣV T =AA ...A Problem: This does not equal Ak Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 7
  • 8. DEDICOM T Solution: A=UXU “DEDICOM – Decomposition into Directed Components” X= Not diagonal Advogato Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 8
  • 9. Computation of Ak with DEDICOM A =UX Uk k T k A is easy to compute Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 9
  • 10. Finding a DEDICOM Singular value decomposition: T A = U(Σ V U ) U T Problem: Not computed to full rank Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 10
  • 11. DEDICOM Algorithms Using singular value decomposition A = U Σ VT LEFT A = U (Σ VT U) UT RIGHT A = V (VT U Σ) VT CLO A = Q X QT U Σ UT + V Σ VT = Q Λ QT (eigenvalue decomp.) X = QT A Q ITER Iterative algorithm (Harshman 1978) (Kliers et al. 1990) Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 11
  • 12. Approximation of eA = I + A + ½A² + ⅙A³ + . . . Approximating A Approximating eA Advogato trust Advogato trust DEDICOM algorithms Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 12
  • 14. References Predicting directed links using nondiagonal matrix decompositions Jérôme Kunegis & Jörg Fliege Int. Conf. on Data Mining, 2012 Models for analysis of asymmetrical relationships among n objects or stimuli Richard A. Harshman Contributions to economic analysis 187, 185–204, 1990 A generalization of Takane's algorithm for DEDICOM Henk A. Kliers, Jos M. ten Berge, Yoshio Takane & Jan de Leeuw Psychometrika 55(1), 151–158, 1990 Jérôme Kunegis & Jörg Fliege Predicting Directed Links using Nondiagonal Matrix Decompositions kunegis@uni-koblenz.de ICDM 2012 14