disassortative degree mixing and information diffusion for overlapping community detection in social...

24
Lehrstuhl Informatik 5 (Information Systems) Prof. Dr. M. Mohsen Shahriari Sebastian Krott Ralf Klamma I5-MS-Monat10- 1 Disassortati ve Degree Mixing and Information Diffusion for Overlapping Community Detection in Social Networks (DMID) Learning Layers Disassortative Degree Mixing and Information Diffusion for Overlapping Community Detection in Social Networks (DMID) Mohsen Shahriari, Sebastian Krott, Ralf Klamma {shahriari, krott, klamma}@dbis.rwth-aachen.de 18.05.2015 Chair of Computer Science 5 RWTH Aachen University

Upload: mohsen-shahriari

Post on 28-Jul-2015

236 views

Category:

Presentations & Public Speaking


1 download

TRANSCRIPT

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-1

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

Disassortative Degree Mixing and Information Diffusion for Overlapping Community Detection in Social Networks (DMID)

Mohsen Shahriari, Sebastian Krott, Ralf Klamma{shahriari, krott, klamma}@dbis.rwth-aachen.de

18.05.2015

Chair of Computer Science 5RWTH Aachen University

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-2

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

Agenda

• Overlapping Community Detection (OCD)• Motivation• DMID• DMID for Time Evolving Networks

• Results• Zachary Karate Club• Evaluation measures and compared algorithms• Synthetic and real-world networks

• Conclusions & Future Works

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-3

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

Overlapping Community Detection (OCD)

Detecting overlapping community structures in networks

Identifying overlapping nodes Two categories of algorithms

- Global approaches [Newman, Mark E. J. and Girvan 2004]

- Local approaches Leader-based methods [Chen et al. 2009; Stanoev et al. 2011]

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-4

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

Motivation

Is the problem solved for all types of networks?- Real world networks have disassortative degree mixing

property Suggesting an algorithm working based on this property

- Competitive with other algorithms Running time Detecting hierarchical structure of graphs Identifying most influential nodes (leaders) Simple logic

- Identifying boundary spanners in learning environments Learning layers project

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-5

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID: A Two Phase Approach

First phase- Identifying most influential nodes- Using of disassortative degree mixing and degree- Identifying local leaders

Second phase- Cascading behavior- Network coordination game

Disassortative network

Assortative network

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-6

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID: Identifying Leaders

Detecting most influential nodes (leaders)- Using of disassortative degree mixing property

- Row normalize disassortative matrix

- Performing a random walk

- Computing local leadership value Combining degree and disassortative value

Cascading behavior named network coordination game

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-7

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID: Identifying leaders

Finding local leaders

Finding leaders using average follower degree (AFD)

ZacharyAFD=8

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-8

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID: Cascading behavior

Network coordination game- Cascades initiated by the identified leaders- Different cascades can overlap- Different cascade size

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-9

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID:Cascading Behavior

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-10

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID:Cascading Behavior

0.5

00.5

0.5

0.33

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-11

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID:Cascading Behavior

1

00.33 1

1 0.5

0.66

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-12

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID:Cascading Behavior

1

00.33 1

1 2

2

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-13

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID:Cascading Behavior

1

01

1 2

2

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-14

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID:Cascading Behavior

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-15

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID for Time Evolving Networks

Using an optimization function for the first phase to detect the leaders - Adding/removing nodes/edges changes the leaders. How to

formulize? Some leaders might be removed (death or merge of communities) Some leaders might be added (split or birth of communities) Leaders do not change (growth or atrophy of communities, continuation) Optimization function?

Detecting cascade changes- How membership of nodes to communities change in each of the

above cases- How cascade sizes change?- How to formulize?

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-16

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

Result on Zachary Karate Club A karate club with 34 nodes and 78 edges Node 1 and 34 as leaders 9, 31, 14, 3, 2 and 20 are overlapping nodes

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-17

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

Evaluation Metrics and Compared Algorithms

Evaluation measures- NMI measure for networks with ground truth communities like LFR

networks [Lancichinetti et al. 2009]

- Extended modularity for evaluation of real-world networks [Nicosia et al. 2009]

Networks for testing and experiments- LFR synthetic networks [Lancichinetti and Fortunato 2009]

- Real world networks Implemented algorithms for comparison

- SSK [Stanoev et al. 2011]

- Clizz [Li et al. 2012]

- MONC [Havemann et al. 2011]

- Link Communities (LC) [Xie et al. 2013]

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-18

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

NMI Measure for LFR Networks

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-19

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

DMID Time Complexity vs Other Algorithms

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-20

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

Results on Real-world Networks

Real-world datasets - Zachary karate club, Dolphin Networks, dblp, Email,

Facebook, Internet, Jazz, Hamsterster, Powergrid, Sawmill, Sawmill Strike

- DMID Highest modularity on Zachary, Sawmill Strike and Internet Second modularity rank on Jazz Best running on time on Email Second running time on dblp, Facebook, Internet and Powergrid DMID is competitive with selected algorithms

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-21

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

Acknowledgement

Funding- Learning Layers Project

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-22

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

Conclusions and Future Works

A two-phase OCD algorithm is proposed- Disassortative degree mixing and information diffusion- Identifies local leaders and hierarchy of the network- Fuzzy membership of nodes to communities- Detecting leaders in social networks- Local nature

Can be implemented distributed

Improving the running time- Implementation with Pregel- Running on huge networks

Experiments on networks with different disassortativity degrees Extending DMID to the case of time evolving networks. How?

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-23

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

REFERENCES Duanbing Chen, Yan Fu, and Mingsheng Shang. 2009b. An efficient algorithm for overlapping community detection

in complex networks. Proceedings Of The 2009 WRI Global Congress On Intelligent Systems (2009), 244–247. DOI:10.1109/GCIS.2009.68

F. Havemann, M. Heinz, A. Struck, and J. Gläser. 2011. Identification of overlapping communities and their hierarchy by locally calculating community-changing resolution levels. J. Stat. Mech. (2011). DOI:10.1088/1742-5468/2011/01/P01023

Andrea Lancichinetti and Santo Fortunato. 2009. Benchmarks for testing community detection algorithms on directed and weighted graphs with overlapping communities. Physical Review E 80, 1 (2009). DOI:10.1103/PhysRevE.80.016118

Andrea Lancichinetti, Santo Fortunato, and János Kertész. 2009. Detecting the overlapping and hierarchical community structure in complex networks. New J Phys 11, 3 (2009), 33015. DOI:10.1088/1367-2630/11/3/033015

H. J. Li, J. Zhang, Z. P. Liu, L. Chen, and X. S. Zhang. 2012. Identifying overlapping communities in social networks using multi-scale local information expansion. Eur Phys J B 85, 6 (2012). DOI:10.1140/epjb/e2012-30015-5

NEWMAN, MARK E. J. AND Michelle Girvan. 2004. Finding and evaluating community structure in networks. Physical Review E 69, 026113 (2004).

V. Nicosia, G. Mangioni, V. Carchiolo, and M. Malgeri. 2009. Extending the definition of modularity to directed graphs with overlapping communities. J. Stat. Mech. 2009, 03 (2009), P03024. DOI:10.1088/1742-5468/2009/03/P03024

Angel Stanoev, Daniel Smilkov, and Ljupco Kocarev. 2011. Identifying communities by influence dynamics in social networks. Physical Review E 84, 4 (2011). DOI:10.1103/PhysRevE.84.046102

Lehrstuhl Informatik 5(Information Systems)

Prof. Dr. M. Jarke

Mohsen Shahriari

Sebastian KrottRalf Klamma

I5-MS-Monat10-24

Disassortative Degree Mixing

and Information Diffusion for Overlapping Community Detection in

Social Networks

(DMID)

Learning Layers

Thank You