Mining Interesting Patterns and Rules in a Time-evolving Graph

Time-evolving graphs, i.e. dynamic networks changing their structures with time, are becoming ubiquitous recently. A typical example is an email communication network whose vertex corresponds to an individual and whose edge corresponds to an email communication within a time period. For the effective analysis of such time-evolving graphs, it must be important to utilize representative local structures in networks as well as time information on edge formation simultaneously. In this paper, we consider a problem of mining frequent patterns and valid rules representing graph evolutions or structural changes in a network with time information. In addition to an effective mechanism for extracting representative patterns and rules, we devise graph-based summarization of discovered rules. By using certain measures provided by the summary, we can expect to find more interesting information that are difficult to obtain in the traditional support and confidence framework. The effectiveness of the proposed framework was confirmed by preliminary experiments using real world email data.