My Photo

« Comparing Google and Microsoft's Mapping Coverage | Main | TailRank, Spinn3r, TechMeme and TechCrunch: New Attention »

October 01, 2007


Adrian Heilbut

Thanks for pointing out this paper - the approach is very nice. The problem of making meaningful visualizations of large graphs is something I've struggled with in the past with protein interaction data. I think that often people end up looking at graphs through exploring local neighbourhoods around one node at a time.

Another idea that this work reminds me of is the PRISM algorithm developed by Segre and Kishony (Nat. Genet, 2004). There, rather than defining classes of nodes a priori, they use characteristics of the graph to define the classes. They are working with graphs with seeral different types of edges, and they look for subsets of nodes that share exactly the same types of edges - that is, they relate to other parts of the graph in a similar way. By partitioning the graph in this manner, they are able to come up with clusters that are functionally meaningful, as well as a visualization in which the nature of the relationships between the clusters is apparent.


This approach looks pretty promising. I always feel a bit jaded when I see another graph visualization, but more exploration functionality seems to really bring the viz out of art/aesthetic land.

The comments to this entry are closed.

Twitter Updates

    follow me on Twitter

    October 2015

    Sun Mon Tue Wed Thu Fri Sat
            1 2 3
    4 5 6 7 8 9 10
    11 12 13 14 15 16 17
    18 19 20 21 22 23 24
    25 26 27 28 29 30 31


    Blog powered by Typepad