Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

The Java Universal Network/Graph Framework (JUNG):

A Brief Tour

Danyel FisherPhD Candidate

Department of InformaticsSchool of Information and Computer Science

with Joshua O’Madadhain and Scott White

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Network Stories

SUNBELT 2000• 270 People & Structural Holes• Database to UCINET to Pajek to MAGE• “I still use UCINET IV from 1992”SOCNET LIST• “I need a new algorithm”• “How does this algorithm do it?”• Reworking it all for a Java applet

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Programmatic Solution

• ProgrammableMake it easy to do the same thing again

• LibraryCan be a server, a client, or a small part of a larger

application

• ExtensibleIf you need a new routine, you can put it in yourself

• Open SourceTake it apart and put it together some other way

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

JUNG

General framework for graph modeling, analysis, and visualization

object-oriented framework for many graphs easy visualization rich attribute structure complex filtering dynamic graphs

Target audience is Java programmers with interest (but not expertise) in graphs

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Overview

• Creating a graph• Annotating with data• Filters• Graph drawing• Algorithms• Demonstration

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Creating a graph

• Load from a file– Pajek– GraphML (XML)

• Build from scratch– Create vertices individually

graph.add( new SparseVertex() );

– Specify edgesgraph.add( new DirectedEdge( v1, v2) );

– (Combine to create SQL or other format reader)

• Use a graph generator– Small-world graphs– Power-law graphs

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Adding user-defined data

key-value pairs associated withgraphs, edges, vertices

easy to write algorithms that operate on relational data

String name = vertex.getUserDatum(NAME);edge.setUserDatum(DATE, new Date());

Decorator infrastructure allows graph to be tagged with standard weights labels indices

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Using graph filters

• Extract sub-graphs of vertices and edges– All edges with user-data “weight” > 0.2– All nodes whose “location” is “California”– All nodes with degree > 1

• Chain filters together– All nodes with “weight” > 0.2

and location is “California” with degree > 1

• Automate complex graph transformations

acceptVertex( Vertex v )acceptEdge( Edge e )

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Simple Graph Display

Quick to display an existing graph.

gd = new GraphDraw(g);jpanel.add(gd);

Renderers draw vertices and edges.

paintEdge( g, Edge, x1, y1, x2, y2 )

paintVertex( g,Vertex, x, y )

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Visualizing graphs

Basic layout algorithms:– Fruchterman-Rheingold– Kamada-Kawaii– Eades– Self-Organizing Maps– Circle

Plug and play architecture- Drop in your favorite renderer, layout

Provided rendering engine is Swing-based - (but layouts are generic)

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Algorithms

clustering (k neighborhood)

connectivitymaximum flownetwork distancesstructural equivalence

centralitybetweennessmarkov

network importance metrics page rankhubs and authorities

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Graph types

now– self-loops, parallel edges– bipartite and k-partite graphs– hypergraphs– mixed graphs (graphs with directed & undirected edges)

coming soon– graph animation

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Statistical Analysis

• Graph measures:– Degree distributions– Average shortest paths– Clustering coefficients

• can use existing Java statistical packages– CERN Colt Scientific Library– Visual Numeric’s JMSL for Java

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Demo

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Conclusion

• Provides a common language for graphs• Complements rather than replaces other graph

packages/network tools• A powerful framework for working with graphs with rich

attribute structure• Ideally suited for building tools/applications related to

network exploration and data mining

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Adoption

• 4000 downloads since August• 200 messages in the forums• Users are writing

– social network analyses– games– animated graph views– trust metrics

Department of Computer Science, University of California, Irvine

Site Visit for UC Irvine KD-D Project, April 21st 2004

Thanks

http://jung.sourceforge.net

Several demos onlineActive collaboration is welcome.