Interactive matching and visual comparison of graphs

1. Interactive matching and visual comparison of graphs MountazHascoët, mountaz@lirmm.fr, LIRMM, CNRS, Univ. Montpellier Pierre Dragicevic, dragice@lri.fr, INRIA, Univ. Paris-Sud Problem Approach Results

2. Lexical network Lexical network for French • « Jeux de mots » • 200 000 Frenchterms 1 200 000 lexical relations ~30 types of relations (hyponymy, hyperonymy, meronymy, associatedideas, etc) ~3 versions of the network eachyear http://www.jeuxdemots.org/ =>Tools to analyze the quality of the network are needed

3. Graph comparison, what for? • Are twosubgraphsdifferent? • Are subgraphs for different relations consistent? • Which graph has lowererror rate?

4. Infovis benchmark • The Infovis 2004 benchmark contains • Metadata for 614 papers • Publishedbetween 1974 and 2004 • Extraction of TopicMaps • Topicsfromtitle and keyword lists • One topicmap per each

5. Graph comparison, what for? • Do severaltopicmapsdiffersignificantly? • How topicsevolve over time • Appear/disappear/refine • How does the connectivityevolve over time?

6. Approach • Optimisation • Use optimizedlayoutfrompreviouswork • Visualization • Introducepresentationstrategiessuitable for graph comparison • Interaction • Leave important matchingchoices to end-users

7. Results • Proposal • Multi-layer model • Animated reconfigurations • Reification of matchingfunctions • Proof of concept • A system named Donatien • Two case studies with real data

8. 1. Multiple layers • Direct Manipulation on layers • Side-by-side arrangement • Superposed • Crossing

9. 2. Animated reconfiguration • Optimizedlayout • Optimisation of aestheticcriteria • Property • Enhancedlegibility of graphs • Limits • If two graphs are the exact same graph • No guaranteethatlayout are the same • Optimizedlayouts • Invariant layout • Deterministlayout • Based on node signatures • Property • 2 identical graphs have the samelayout • Warning: reciprocally not true • Limits • Legibility of graphs

10. Visual comparisonthrough reconfiguration

11. Comparisonwithmatching • Requirements • Computematchingnodes • Interactivelyadjustthismatching • Superimposematchingnodes • Solution • Create a master graph • From invariant layout • To embody the matchingfunction in a layer • Integratedsmoothlywithotherlayers

12. 3. ReifyingMatchingFunction • Notations • G1(V1,E1), G2(V2,E2)… Gn(Vn,En) • graphs to becompared • M : A matchingfunction • A hypergraph H(V1uV2u…Vn,EM) • Master Graph • Embodies the MatchingFunction • G(V,E) is a graph • V represents hyper-edges of H • n  V, (v1,v2,…vk)  EM • n -> (v1,v2,…vk) • E representsedges of E1, E2, … En

13. ReifyingMatchingFunction

14. MatchingExample

15. Conclusion • Visual comparison of graphs • Invariant lMulti-layer model • Animated reconfiguration • ReifiesMatchingFunction • Layout and superposition • Smoothintegration of master graph • Future work • Enhance interaction model • Enhance invariant layouts

16. Thanks for your attention demoat lunch Questions ?

17. Comparingclustered graphs