Dynamic Community Detection with Normal Distribution in Temporal - - PowerPoint PPT Presentation

▶
dynamic community detection with normal distribution in
SMART_READER_LITE
LIVE PREVIEW

Dynamic Community Detection with Normal Distribution in Temporal - - PowerPoint PPT Presentation

Jan 04, 2023 372 likes •666 views

Dynamic Community Detection with Normal Distribution in Temporal Social Networks Yaowei Huang Yuchen Lin Zhaozhe Song 5140309539 5140309507 5140309514 14330222150355@sjtu.edu.cn yuchenlin@sjtu.edu.cn zhaozhesong@sjtu.edu.cn Supervisor:

slide-1
SLIDE 1

Dynamic Community Detection with Normal Distribution in Temporal Social Networks

Yaowei Huang 5140309539 14330222150355@sjtu.edu.cn

Yuchen Lin 5140309507 yuchenlin@sjtu.edu.cn Zhaozhe Song 5140309514 zhaozhesong@sjtu.edu.cn

May 2017

Supervisor: Prof. Luoyi Fu

slide-2
SLIDE 2

Evaluation and Simulation

宋肇哲

slide-3
SLIDE 3

Evaluation and Simulation

  • Novel. Have to design some metrics by ourselves.
slide-4
SLIDE 4

Example

slide-5
SLIDE 5

Example Values of F

F1:13.5 F2:2.4 F1:4.2 F2:9.3 F1:0.3 F2:11.0 F1:0.5 F2:10.6 F1:0.1 F2:19.6 F1:9.2 F2:0.2 F1:9.7 F2:0.4 F1:9.6 F2:0.3

slide-6
SLIDE 6

Example Values of μ

2000 2004 2005 2007 2004 2012 2011 2008

slide-7
SLIDE 7

But, how can we evaluate the results quantitatively?

slide-8
SLIDE 8

Two aspects

  • The community weight (F)
  • The temporal dimension (μ , σ)
slide-9
SLIDE 9

Evaluation on the community weight F

  • Average F1 Score
  • Omega index
  • Accuracy in the number of communities
slide-10
SLIDE 10

Ground truth: Community number

Problem:

But our detected result…. Only anonymous communities

slide-11
SLIDE 11

Find the most similar matching for each community!

slide-12
SLIDE 12

Ground truth: Community number

Problem:

But our detected result…. Only anonymous communities

slide-13
SLIDE 13
slide-14
SLIDE 14

Evaluation on the community weight F

  • Average F1 Score
  • Omega index
  • Accuracy in the number of communities
slide-15
SLIDE 15

Average over all detected and ground truth communities: The best matching for our detected result The best matching for ground truth *Note: not one-to-one matching

slide-16
SLIDE 16

Evaluation on the community weight F

  • Average F1 Score
  • Omega index
  • Accuracy in the number of communities
slide-17
SLIDE 17

Omega index

estimating the number of communities that each pair of nodes shares 


slide-18
SLIDE 18

Evaluation on the community weight F

  • Average F1 Score
  • Omega index
  • Accuracy in the number of communities
slide-19
SLIDE 19

Accuracy in the number of communities

slide-20
SLIDE 20

Evaluation on the community weight F

  • Average F1 Score
  • Omega index
  • Accuracy in the number of communities
slide-21
SLIDE 21

Some baseline methods do not scale well. Solution: Sample subnetworks

  • pick a random node u that belongs to at least two communities
  • pick all the nodes that share at least one same community with u
slide-22
SLIDE 22
slide-23
SLIDE 23

Two aspects

  • The community weight (F)
  • The temporal dimension (μ , σ)
slide-24
SLIDE 24
  • Pearson Correlation

Evaluation on the estimated temporal factors

(μ , σ)

Time Membership Strength

Ground Truth

Membership Strength Detected Strength Distribution

Pearson Correlation

slide-25
SLIDE 25
slide-26
SLIDE 26

Challenges

  • Dataset too large
  • Fitting process very slow
  • May suffer from local minimum
slide-27
SLIDE 27

Future improvement

  • Improve gradient ascent algorithm for

faster speed

  • Find better smaller datasets
  • Use normalization or regularization for the

parameters

slide-28
SLIDE 28

Thank You!