90 likes | 221 Views
Node-Attribute Graph Layout for Small-World Networks. Helen Gibson Joe Faith. IV2011 - AGT. Intelligent Modelling Lab. Small-World Networks. What are they?. Clustered with a high clustering coefficient Smaller than average shortest path length. Examples. Milgram (1967) IMDB. Layout.
E N D
Node-Attribute Graph Layout for Small-World Networks Helen Gibson Joe Faith IV2011 - AGT Intelligent Modelling Lab
Small-World Networks What are they? • Clustered with a high clustering coefficient • Smaller than average shortest path length Examples • Milgram (1967) • IMDB Layout • Force Directed • Packed together • Lose clusters • Users IV2011 - AGT Intelligent Modelling Lab
Node-Attributes Information about the nodes • Retinal Variables • Colour • Size • Shape What about nodes having multiple classifications? Or lots of quantitative attributes? IV2011 - AGT Intelligent Modelling Lab
Node Attributes Attributes (O) Nodes (X) IV2011 - AGT Intelligent Modelling Lab
Dimension Reduction + TPP Targeted Projection Pursuit – interactive high-dimensional data exploration J. Faith, “Targeted Projection Pursuit for Interactive Exploration of High- Dimensional Data Sets, ” 11th International Conference Information Visualization (IV ’07), Jul. 2007, pp. 286-292. IV2011 - AGT Intelligent Modelling Lab
Example Application Gephi Randomly add and remove links • Force-directed • Yifan Hu Clustered Remove attributes Assign attributes gephi.org IV2011 - AGT Intelligent Modelling Lab
Example Application Targeted Projection Pursuit • Attributes as dimensions • Number of attributes = • Number of dimensions Which attributes are significant in clustering? http://code.google.com/p/targeted-projection-pursuit/ IV2011 - AGT Intelligent Modelling Lab
Example Application LinLog - Andreas Noack (2007) • Energy Models • Force Directed • Graph Clusterings http://code.google.com/p/linloglayout/ IV2011 - AGT Intelligent Modelling Lab
Conclusions + Further Work • TPP - greater visual separation than force-directed layout • TPP – doesn’t lose the context that LinLog does But… • Further empirical validation needed! • Metrics • Vary parameters • Insights gained • Further use of attributes Most importantly… • Real world applications http://code.google.com/p/targeted-projection-pursuit/ IV2011 - AGT Intelligent Modelling Lab