Spotting communities within networks is a big deal. Not least for search engines that rely heavily for their results on the communities that form when websites point to each other. If a lot of websites point to another site then that proves it is of value.
At least that’s what everyone has assumed. But links can be negative as well as positive. If lots of websites point to another site specifically to say how bad it is, then the community is actually saying the site has little value.
Being able to tell the difference, then, is crucial, not only for search results but in understanding the structure of the network and the communities that emerge.
Vincent Traag at the University of Amsterdam in the Netherlands and a buddie say that including negative as well as positive links, profoundly changes the pattern of communities that you find in a network.
They’ve applied the idea to a dataset called the Correlates of War that provides details of agreements and disputes between 138 countries between 1993 and 2001.
In terms of the network, a negative link is the same as a positive link but pointing in the opposite direction (it has the opposite sign).
By putting the links into a model of the world, Traag has worked out what global communities existed at the time. The communities that emerge are the standard power blocs well known to historians: the West; Latin America; Russia & China; West Africa; North Africa & the Middle East; and a collection of independents not truly forming a bloc.
That’s almost exactly as historians would put it except for one or two features. For example, west Africa does not normally figure as a power bloc on its own and the independents include New Zealand which would normally be classified as part of the West.
That provides an interesting and somewhat unconventional insight ino the politics of the time.
Ref: arxiv.org/abs/0811.2329: Community Detection in Networks with Positive and Negative Links
This is interesting work. Worthwhile getting a designer to redo the graphic, around the countries’ capitals instead of their centroids.
At a higher level, it reminds me of Notes on the Synthesis of Form by Christopher Alexander (I’m surprised he wasn’t in the paper). In the book, he defines a “fitness graph” as a tool to decompose complex design problems. The graph has positive and negative arcs between nodes, respectively representing corroborating or conflicting requirements. For his purpose, he collapsed the arcs into a single type: “informs”, but I always thought the distinction was of particular interest.
Its really interesting how these ‘organizations’ take shapes due to positive or negative linking. As in a society, linking is vital in neural and web applications too.