2010 年 25 巻 3 号 p. 452-463
Most existing works on network analysis mainly focus on only the existence of relations between entities. However, in trying to understand a real network, we naturally use not only the existence of relations but also information on the kind of relations, the attributes in the nodes, and the changes in time. In addition, we can observe some of the measures that are obtained as a result of the whole network structure. In order to extract some meaningful structural changes and integrity constraints from a dynamical network constructed from survey data, we are proposing a novel data mining framework in this paper that includes the above information that has not been used in previous studies. In the proposed framework, we start by detecting the change point in the dynamic network according to the change in the characteristic quantity. Then, by using the detected points, a dynamic network will be divided into two groups. In other words, we associate the class information to each network in a dynamic network. Finally, meaningful structural changes and integrity constraints can be obtained by applying inductive logic programming to a dynamic network and the related background knowledge represented in the first order logic. In experiments using real world data, we succeeded in obtaining meaningful results. Thus, we confirmed the usefulness of the proposed framework.