Open Access News

News from the open access movement

Monday, February 16, 2004

Butterfly effect in OA eprint repositories

Jon Kleinberg, Cornell professor of computer science, is studying the propagation of influence by studying email lists, blog networks, and the arXiv repository of scientific eprints. Excerpt from the Cornell press release: "The researchers tested their algorithm [for identifying influential people] on another kind of network, the pattern of co-authorship in scientific papers. Their data pool was the online E-print Archive of physics and mathematics publications, commonly known as the arXiv, maintained by Cornell University Library. People were considered to be linked when they co-authored papers. The studies ignored any real-world information, such as whether two people might be at the same institution. In simulations Kleinberg and colleagues found that their method significantly outperformed methods that rely solely on counting links or measuring the distance between candidates and the rest of the network. Kleinberg also has been studying the way networks grow over time, working with David Liben-Nowell, a Ph.D. student at the Massachusetts Institute of Technology. One goal is to try to predict where new links will form in a network. In the arXiv network, the researchers hypothesized that two people who haven't been linked would be likely to form a link if they are near one another in linkage terms. What they found, however, was that the number of hops was not the best measure of nearness. The reason, Kleinberg says, is the 'small world phenomenon' -- the fact that everyone is on average 'six degrees of separation' from everyone else -- so counting the number of hops between people doesn't help. 'It's better to look for people who have many different short paths connecting them,' he says. 'This is an interesting open question with a lot of room for further research.'" (PS: Yes, and this research will be relevant to building new and better metrics for the impact of OA research articles.)