1 / 15

Topologically biased random walks with application for community finding

Vinko Zlatić Dep. Of Physics, “Sapienza”, Roma, Italia Theoretical Physics Division, Institute “Ruđer Bošković”, Z agreb, Croatia. Topologically biased random walks with application for community finding. CNR-INFM Centro SMC Dipartimento di Fisica, Universita di Roma “Sapienza”, Italy.

Download Presentation

Topologically biased random walks with application for community finding

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Vinko Zlatić Dep. Of Physics, “Sapienza”, Roma, Italia Theoretical Physics Division, Institute “Ruđer Bošković”, Zagreb, Croatia Topologically biased random walks with application for community finding CNR-INFM Centro SMC Dipartimento di Fisica, Universita di Roma “Sapienza”, Italy Andrea Gabrielli Guido Caldarelli Zlatić, Gabrielli, Caldarelli, arXiv:1003.1883

  2. Community Finding Community: One of many (possibly overlapping) subgraphs Has strong internal node-node connections Weaker external connections Q:How to find communities in large networks? Santo Fortunato “Community detection in graphs”, Eprint arXiv: 0906.0612, Physics Reports 486, 75-174 (2010)

  3. Motivation Distribution of links in the network is locally heterogeneous 1. Finding sets of nodes with similar function 2. Visualization 3. Classification 4. Hierarchical organization Useful in: Systems Biology, Sociology, Computer sciences, Physics ...

  4. Community Finding(2) Algorithms:S.Fortunato, arXiv:0906.0612, Physics Reports 486, (2010) pp.75 • Newman-Girvan, removal of links with high beetweeness • Different algorithms to maximise Newman modularity • Radicchi et al. ,removal of links based on local properties • Cfinder, Markov cluster algorithm,Potts model ...

  5. Spectral Methods • Idea: distance in the M dimensional space spanned by eigenvectors associated with random walks • M corresponds to number of eigenvectors used • Then we can apply standard clustering techniques Manhattan distance, angle distance, etc • Graph Laplacian L. Donetti and M. A. Munoz, J. Stat. Mech. P10012 (2004).

  6. Problem Community structure can be very hard to detect Mixing Parameter Different performances of algorithms (detectability, speed, size of networks) A. Lancichinetti, S. Fortunato, Phys. Rev. E, 80, 056117 (2009). Donetti and Munoz is one of better algorithms

  7. Biased random walks IDEA • New idea: Different topological quantities have different frequencies in between different communities Why not use this additional information to improve spectral methods??? Example: Edge multiplicity Other possibilities:Shortest path betweeness, subgraph frequencies, degree, clustering, even eigenvectors of nonbiased random walk.

  8. Biased Random walks Unbiased transition operator (Frobenius-Perron operator) Biased transition operator (Frobenius-Perron operator) Exponential familly of transition probabilities Asymptotic probabilities Detailed balance condition

  9. Symmetrization(1) Symmetrization leeds to hermitian operator. Orthogonal eigenvectors

  10. Symmetrization(2) Parametric equations of motion D. A. Mazziotti, etal, Journal of Physical Chemistry 99, 112-117 (1995).

  11. Spectra Spectra contains important information on community structure N separate graphs considered as one have N-fold degeneracy of the first eigenvector Characteristic time to approach stationary distribution is related to spectral gap N communities should produce N-1 close eigenvalues!

  12. Application to community finding(1) Indeed!!! Close large eigenvalues form clear “community band” for modest values of mixing parameter There is a clear separation between N-1 eigenvalues associated with community structure and rest of the spectra Separation beetween Nth eigenvalue and N+1th eigenvalue we name “community gap” Tuning of biases in such a way to maximize the community gap

  13. Application to community finding(2) Every network has its own optimal parameters Tetrahedric structure – description with angles

  14. Conclusion 1.This topic is still “hot” 2. Possibility to include variables related to dynamics on networks 3. Promissing preliminary results (outperform DM 0.60 vs. 0.45 at mixing parameter =0.5) To do: Test different topological variables as basis for biases Develop better clustering algorithm based on distances between nodes.

  15. Thanks for your attention

More Related