Science News Highlights Research on Influence and Strategy in Social Networks
|Go Back to Home|
Science News, a popular-media bi-weekly magazine of the Society for Science and the Public, has recently reported, in its June 8th web edition (and in an upcoming paper edition), on research performed by Luis Ortiz, assistant professor in the Department of Computer Science at Stony Brook, and his research group. Mohammad Irfan, a graduate student member of the research group, presented the research results at the Workshop on Information and Decision in Social Networks held at MIT May 31-June 1st:
Mohammad T. Irfan and Luis E. Ortiz. A model of strategic behavior in networks of influence. In WIDS´11.
The research results, highlighted in the Science News report under the heading "Power networks in Congress," is part of a larger research project being carried out in Ortiz´s group on computational game-theoretic approaches to the study of networks and behavior. The specific project mentioned in the report concerns the application of computational game theory to the study of influence in very large populations of individuals in complex networks of interactions. Through this research, members of Ortiz's group designed and proposed a game-theoretic model of behavior they call "influence games."
One application of the model is to identify the most influential individuals in a complex social network. In contrast to previous approaches, the researchers defined "most influential" using strictly the concept of "Nash equilibrium," so called in deference to Nobel laureate John Nash´s seminal work back in 1951 establishing its universal existence. One instantiation of the general approach addresses situations in which one seeks to achieve or induce homogeneous behavior known to be stable in the system by identifying the smallest group of individuals whose adoption of the behavior would "force" all other individuals to also adopt the behavior, thus leading to the desired homogeneous state. As a proof of concept, the researchers applied their approach to identifying influential U.S. Senators from congressional voting data. The influence game models were inferred from real voting data from various sessions of U. S. Congresses using more general machine-learning techniques, specially designed for that purpose in collaboration with Jean Honorio, a graduate student associated to the group.
The news brief makes an intriguing connection to the current "gang of six" U.S. senators trying to work out an agreement on the government´s budget in an attempt to cut public debt. It is curious that the analysis carried out by the researchers using the proposed approach also suggests that there were six most-influential senators (3 republicans and 3 democrats) in the previous congress. According to their model, if the identified senators decide to vote "yes" on a legislation, then all other senators would also vote "yes." A more detailed version of the research will be presented in this year´s Conference of the Association for the Advancement of Artificial Intelligence (AAAI) to be held in San Francisco in August 2011:
Mohammad T. Irfan and Luis E. Ortiz. A game-theoretic approach to influence in networks. In AAAI´11.