Yesterday concluded the first annual Workshop on Information in Networks at NYU’s Stern business school. Though I arrived late and left early both days—blame it on the conference being on my “home turf” and thus me not being able to ignore local commitments—I took a lot away from the conference, here are my thoughts:
- Despite the explosion of interest in networks, the research community is still rather small, and 90% of it was at WIN. This, of course, was a great achievement for the conference, but it also reinforced the cyclical intellectual growth of the community. That is, most of the seats were occupied by high profile network researchers, who listened to talks by their colleagues on research most people were already familiar with. As a result, there were very few attendees bringing an outside perspective or walking away with a new appreciation for the discipline. Ron Burt may have had the quote of the weekend with (paraphrasing),”we all came here wearing full body condoms, not open to new ideas.” There is, however, easy ways to fix this: advertise better, get a bigger room, and provide young (less known) scholars with speaking time rather than just poster sessions.
- On a somewhat related point, giving “high profile” scholars a platform is great, but there must be some conditions on the—let’s say—completeness of what they are presenting. Though they were certainly the exception rather than the rule, a few presenters were particularly egregious in this respect. I realize this is a systemic problems for most academic conferences, but the network community seems to have more than its fair share of “hero worship”. Put these big names in the audience and allow them to provide feedback for young scholars; much more value in that.
- Pedro Domingos presented a fascinating paper on the development of an appropriate language for network science. Domingos terms his language Markov logic networks (MLN) as a means to describe network processes in an analytical way. This is a subject of great interest to me, as I have previously discussed the need the need to develop a language (or mathematics) to describe network dynamics. The framework presented by Domingos is compelling, though in some ways it appears to be attempting to fit all casual problems into a networked logical structure, which is useful, but does not address the problem of understanding network trajectory. I am curious, however,if MLN can be developed to support this type of analysis. If so, it is a very exciting breakthrough.
- Beyond the obvious reasons for not showing up late to conferences, a really good reason to arrive early is to secure a good seat. Being late also meant that I was banished to the extremities of the room, which among other things meant session moderators never saw my hand go up. As such, here are a few criticisms that I was not able voice during the conference:
- Albert-László Barabasi presented a detailed talk on his current research on the travel patterns and predictability of humans. The finding Barabasi was eager to highlight was that human deviations from travel patterns does not appear to increase with the average distance traveled. He posited that this was counter-intuitive because people who normally travel large distances should have a great opportunity to deviate. This seems reasonable, however, what was missing from the presentation was an accounting for the increased constriction the physical network of roads, bridges, waterways, etc. has on travel as distance increasing. My intuition is most of this finding can be explained through including this into the analysis.
- Arun Sundararajan presented his research on shock diffusion through networks; a topic of great interest to me. The research investigates how exogenous shocks, such as having a New York Times review or being featured by Oprah, affect the demand for books sold online, and then how these shocks are absorbed by second, third and forth order “related” books as implemented by Amazon.com. Sundararajan noted the surprising statistical result that very often third and forth order books maintain increased levels of demand after these shocks. What he did not consider, however, is the possibility that these shocks are not simply raising increasing the visibility of a book or an author, but in some cases they are raising the salience for a broad issue. For example, Sundararajan noted how after Jenny McCarthy’s book on being the parent of an autistic child was featured on Oprah it induced these distance shocks for related books, but a competing perspective could be that this event was part of a broader shock to the overall salience of autism. One way to control for this would be to include a dummy variable for fiction or non-fiction books. Non-fiction books will have a higher propensity for this salience effect; therefore, this would be a simple way to address this question.
Update: From Prof. Sundararajan in the comments, “This probably didn’t come across well at the talk: but we do consider the effects of an increase in visibility of both the topic and of the author, and use controls for both (category and same-author); the results are robust to these controls.”
Finally, it was great to finally meet Daniel Katz and Jon Zelner (Jon even schooled me on Python multi-threading, which I am eager to implement) from the University of Michigan political science department. These guys do very interesting work on Supreme Court citations networks, and their poster was one of the most popular at session. I highly recommend following their work.
Photo: Workshop on Information in Networks
Automatically Generated Related posts:
Hi,
Must have been an interesting meeting. Thanks for the notes.
I saw a talk of Katz & Zelner on the Supreme Court citation networks on the conference in Lisbon last year, october if i remember correctly. It was indeed by far one of the most interesting talks there.
[Reply]
Drew, thanks for your great summary and thoughts.
This probably didn’t come across well at the talk: but we do consider the effects of an increase in visibility of both the topic and of the author, and use controls for both (category and same-author); the results are robust to these controls.
[Reply]
I enjoyed the conference and learning from it. But, 90% of network analysts were NOT at conference (rather than the 90% you say were at it). There are huge worlds out there. This was a pretty inbred group. Come to the Sunbelt Social Network Conference, Lago di Garda, late June 2010, or read the journal Social Networks. Accessible via http://www.insna.org. And of course, the Socnet email list.
[Reply]
Arun, thanks for the clarification. I have moved it up into the main body of the post.
Barry,
First, as a point of pedantic order, I do say that it was 90% of the “network community,” not network analysts as you state. You have made distinction between network analysis and network science, so it seems important that we remain consistent. To the point, I suppose it is about perspective, i.e. you have one view of who is a network analyst, I may have another (I wonder if YOU think I am one?). The number 90% was used to make the point, that you then repeat in the comments, which is it was a very inbred gathering; somewhat to its detriment. Regarding Sunbelt, I have had papers accepted, but as poor grad student I have never been able to afford grand escapes like Lago di Garda, Italy. Of course, if I can work out timing and funding I will be there—bells affixed.
[Reply]