en-de
en-es
en-fr
en-sl
en
en-zh
0.25
0.5
0.75
1.25
1.5
1.75
2
Classification and Clustering in Large Complex Networks
Published on 2011-06-148178 Views
Data represented as graphs with complex structures are ubiquitous nowadays. Examples include technological networks (e.g., the Internet), informational networks (e.g., the World-Wide Web), social n
Related categories
Presentation
Classification and Clustering in Large Complex Networks00:00
Wordle TM says ...00:07
Complex Networks are Ubiquitous00:59
Problems / Applications (1)02:11
Outline (1)04:26
Relational Classifiers04:39
Within-Network Classification05:44
Background: Previous Work on Within-Network Classification06:45
Our Work Addresses Two Challenges for Within-Network Classification07:43
Limitations of Existing Approaches for Within-Network Classification08:12
Our Solution: Ghost Edge Classifiers Have Consistently High Performance10:46
Challenge 1: Label Sparsity - The Standard Approach11:26
Challenge 1: Label Sparsity - The Ghost Edge Approach12:12
Challenge 1: Label Sparsity - Weighting Ghost Edges (1)13:26
Challenge 1: Label Sparsity - Weighting Ghost Edges (2)13:50
Challenge 2: Non-homphily - How to handle degrees of homophily? (1)14:32
Challenge 2: Non-homphily - How to handle degrees of homophily? (2)14:39
The GhostEdge Classifiers16:54
Summary of Data Sets & Prediction Tasks19:50
Various Ways of Measuring Homophily on a Network21:20
Dyadicity and Heterophilicity22:34
Log Odds of Dyadicity & Heterophilicity24:18
Experimental Methodology for Evaluating Within-Network Classifiers25:39
We Ran Seven Individual Classifiers26:33
GhostEdgeNL is Top Performer Among Relational Neighbor Classifiers27:59
GhostEdgeL is Top Performer Among Link-Based Classifiers28:49
Summary: Ghost Edges29:22
Outline (2)31:18
Community Discovery31:47
Measures of Effectiveness33:39
Background35:19
Modularity (FM)38:22
Cross-Associations (XA)38:42
Latent Dirichlet Allocation for Graphs (LDA-G)39:00
LDA-G, FM, and XA are not Consistent w.r.t. Link Prediction40:15
Hybrid Community Detection Framework (HCDF)41:02
More Whitespace, Higher Average AUC on Link Prediction42:16
HCDF with Attribute Coalescing Strategy44:32
Plate Model for HCDF with Attribute Coalescing Strategy45:18
HCD Algorithm45:35
Summary of Real-World Graphs Used in Experiments (1)45:50
Summary of Real-World Graphs Used in Experiments (2)46:15
Summary of Real-World Graphs Used in Experiments (3)46:35
Measuring Effectiveness Quantitatively: Link Prediction (1)47:25
Measuring Effectiveness Quantitatively: Link Prediction (2)48:28
Experimental Methodology48:53
Link Prediction Performance Across Various Graphs48:59
Worst-Case Link Prediction Performance Across All Graphs50:32
Measuring Effectiveness Quantitatively: Value of Information51:17
Robustness Measured across Various Real-World Graphs52:11
What’s Going on Here?53:13
What About a Super-Hybrid?54:06
Summary: Community Discovery55:07
Conclusions56:02
Thank You56:45