CitationNetworkVisualization
From ImageWeb
CITATION NETWORK VISUALIZATION
Naive notes from David, 17 July 2008
A word of caution: "Because motion draws the eye so well, it is very important that nodes move in a graph only when they undergo a significant structural change, in order for the animation to convey useful information."
Contents |
Biological analogies
- Here are two biological analogies of how citation networks grow year-on-year, as new papers cite existing ones, that we might collectively wish to brainstorm:
1. The new citations add a new layer of links on top of the existing three-dimensional citation network (in which the z dimension is time). This is not a million miles away from Alistair's concept of growth in the Garden of Meaning, in which spiders spin new links, the topmost layer of links being the most recent "take" on the world (although I clearly realise the difference between the evolution of a single ontology in Al's model and the network of citations between multiple distinct papers):
"There are also spiders in the vineyard, and gardeners can induce these spiders to spin silk threads of different types between stem nodes. The superstructure created by all of the threads in a vineyard is the spiders' web. The current state of a conceptualisation is given by the terminal stem nodes, or tips, in the vineyard, and the threads between them – the part of the spiders' web connecting only tips to other tips is known as the web surface, an “aerial photograph” of the vineyard."
2. A paper that reaches out and makes a new citation link to a paper in another discipline area, at the start of the growth of a new interdisciplinary area, can be likened to one of Mark Fricker's fungi or slime moulds, that dynamically creates a mycelial link to a new distant food source, weak at first, then strengthening if the interaction is beneficial.
Use of Protege-OWL to infer citation networks
- I had the crazy idea of using Protege-OWL to record, not hierarchies of ontology classes, but citations of publication instances. If you entered the citations of each paper as a separate asserted hierarchy of depth one, you could then infer the complete citation network from identities of citations between individual trees, and then display the inferred citation network. Would that work, permitting us to use a reasoner to infer the network and the OWLViz tool to visualise it? The problem with this is that the visualization tool doesn't work in 3D.
Network visualization incorporating temporal display
Blythe and Oats "[Visualization support for fusing relational, spatio-temporal data, http://maple.cs.umbc.edu/papers/fusion06-blythe.pdf]" do this just in 2D (their Fig. 2). Furthermore, they simplify display of a citation network by displaying not citation links, but co-citation links, and by omitting links representing 4 or fewer co-citations. This is a cunning concept, quite different from my ideas of citation link weighting and typing: If Paper A and Paper B both cite Papers C, D, E, F and G, a co-citation link is shown between A and B. Papers heavily linked by co-citation links cluster into domains of interest. Since co-citations will have a strong temporal similarity, I am unable to figure out how how a co-citation network does longitudinal linking between years (thus relating Papers A and B to Papers C, D, E, F and G). The Blythe and Oats paper, which confusingly mixes papers and authors in the same graph, does not make this clear.
Blythe and Oats have a number of other insightful ideas, including the use of flow and wave dynamics, and the use of difference graphs that show only the change between situation/time A and situation/time B.
The KrackPlot program used by Blythe and Oats has the great advantage that the nodes can be geolocated onto maps ("Because the latitudes and longitudes of nodes are given, KrackPlot can lay the network over a map, given as an image with its corners registered. When nodes are positioned on a map by the locations of the reports, this type of display can show patterns of reports (e.g. of SARS infections) evolving through space and time,").
The network model at http://cmol.nbi.dk/models/infoflow/infoflow.html has a nice colour-coding for the age of information. This could be exploited for citation analyses, if one did not want to use time as a graph axis.
Network visualization software
KrackPlot
The KrackPlot program used by Blythe and Oats is downloadable from [here, http://www.isi.edu/~blythe/KP/setup-KP4.3.exe]. Worryingly, the current version of KrackPlot is 4.3, built on 11/02/2006. Elsewhere it is said to be obsolete, superseded by Pajek.
Pajek
Pajek (http://pajek.imfm.si/doku.php?id=pajek) is currently supported and has a wiki (http://pajek.imfm.si/doku.php). If you go to the Pajek site, you will see why this might be appropriate for the SPIDER Project!
Other links
NetVis (http://www.netvis.org/index.php) is another free visualization package, but I have not checked it out.
One of the most innovative visualizations for hierarchical networks I have come across is the botanical one at http://www.win.tue.nl/~vanwijk/botatree.pdf.
And one of the prettiest visualizations is of the nested self-similarity structure within a piece of music: http://www.turbulence.org/Works/song/gallery/gallery.html.
Mark Rounds of Qinetiq has a current blog on (social) network analysis: http://mdround.blogs.com/.
Netwiki (http://netwiki.amath.unc.edu/) is a wiki concerned with network theory.
There is a Wikipedia entry at http://en.wikipedia.org/wiki/Social_network, particularly defining key concepts.
