Graph mining has been studied in many fields for obtaining useful information from a graph by finding its characteristic substructures. For example, several methods for Web graph mining has been proposed. Some of these methods identify communities on the Web by finding complete bipartite graphs. Most methods for Web graph mining deal with only static graph structures whose nodes are pages and edges are links, that is, they utilize no time-series data of pages and links, such as the creation date of each page or link. The Web is dynamic, however, and the graph structure of the Web is changed dynamically over time. Therefore, it is expected that we can obtain more useful information by utilizing time-series data of pages and links.

We propose a new technique for Web graph mining to utilize time-series data of pages and links, by introducing a node label reflecting the creation date of the page corresponding to the node. By applying our technique to graphs among blogs and news sites, we succeeded in identifying some characteristic patterns in them.