From Santa Fe Institute Events Wiki
Statistical Inference for Complex Networks Workshop, December 3-5, 2008, Santa Fe NM
Mark Newman (homepage)
Identifying types of nodes in a network
Suppose the nodes in a network are of a number of different classes: the individuals in a social network have different interests or speak different languages; pages on the web are on different topics and are written in different languages; nodes in a metabolic network are divided among the different mechanisms in the cell, and so forth. If we have metadata describing the roles of nodes then it may be easy to determine these classes, but what happens if we have only the network topology to go on? This talk describes a method for classifying network nodes according to similarities in their patterns of connection to other nodes. The method, which makes use of an expectation-maximization algorithm, is able not only to assign nodes to classes, but also to determine dynamically the best definition of the classes so as to capture in the classification the maximum amount of information about the network structure. We give a number of example applications, showing how the method is able to detect unusual but useful forms of structure in networks taken from a variety of areas.
Joint work with Elizabeth Leicht (UC Davis).