Monthly Archive for February, 2011

My Sunbelt Top 7

Last week I was lucky enough to get to INSNA’s Sunbelt conference in Florida. Here’s my top seven papers:
  1. Kevin Lewis presented some work done with Andrew Papachristos on the structure of gang warfare in Chicago using data on inter-gang murders.  Kevin described putting a stronger methodology on data from an earlier paper (pdf). One thing I loved about it was the mapping of street terms to abstract network structures – ‘payback’ = reciprocity, ‘untouchables’ = high out-degree, low in-degree, etc.
  2. Jamie F Olson talked about the statistical properties of centrality measures of communication networks over time. I didn’t quite grok the talk but the gist was that by varying the time-window size and comparing centralities across time periods it is possible to identify the ‘best’ sampling window for the network. For example, he showed that a week was a good period to sample some email data. Apparently a preprint may be available soon on his personal page – I’ll be looking out for that!
  3. Ulrik Brandes (with Bobo Nick) gave a beautifully crafted visualisation design paper. They used Gestalt principles to put together sparklines-inspired glyph for showing network dynamics.  Very elegant.
  4. Elisha Peterson had some smart ideas for keeping node positions stable in visualisations of dynamic networks. He did this by putting springs between versions same node across the time slices (before & after). It seemed to make things more stable at the expense of some calculational complexity.
  5. Lin Freeman shared his insights on the many ways of finding cohesive sub-groups in networks. He gave a clear and concise history of various methods from social sciences, maths & physics. Then an outline of measures of success (modularity q, EI conductance, Freeman Segregation index, Pearson’s correlation ratio) before running the algorithms over a collection of data sets.  Success depends not only on the algorithm but also of course on the cohesiveness of the data. Conclusion?  Good:  Correspondance Analysis, Leading Eigenvector, WalkTrap, Fast Greedy. Not so good: Factions, Tabu, others.  I hope this work gets written up in a review paper soon.
  6. Mark Lauchs talked about the networks involved in a massive police corruption case in Queensland, Australia that were exposed by the Fitzgerald Inquiry. This talk demonstrated that it probable that ‘dark networks’ can never be found automatically: the bad cops were structurally similar to the good cops. The only practical way of uncovering the network inside is to identify at least one bad egg, and use network structures to work from there to get the wider picture.
  7. Joshua Marineau had some interesting insight into the benefit of negative ties within an organisation.  Although it has been shown that individuals who have negative ties under-perform, he claimed that being positively connected to someone who themselves have negative ties can actually be an advantage.
The legendary hospitality suite was as friendly as ever too ;-)

The Birth of a Link

This diagram is absolutely fascinating.
It comes from Easley & Kleinberg’s new book from an excellent paper by Crandall et al (2008) (pdf).
It is a sort of anatomy of how links between people are created: it tries to capture the birth moment and the forces before and after it.
The upward curve is intriguing but straightforward to explain by homophily – like seeking like.
The most interesting bit is the curve just before the first communication occurs.  People get suddenly more similar – a kind of gravitational attraction occurs in the affiliation network and the first communication is sparked into life, closing the triads.
Although is tempting to explain this by creating physics based models, as the paper does,  I can’t help feeling there is a simpler explaination.   I would guess that the base of the curve is generally where ‘awareness’ happens.  At this moment the editors become aware of each other, and at that point a basic psychological effect takes over: simple curiosity. People actively seek each other out, viewing each other’s activities and building a picture of the type of the other person. Partly this is also to de-risk the first encounter in order to make the right first impression.
It isn’t often that one sees abstract concepts like curiosity in science, but I guess that is the power of big data & a great set of research questions ;-)