Exploring graph mining approaches for dynamic heterogeneous networks

Abstract

As we become a more 'connected' society, a greater need exists to understand complex network structures. While many in the field of data mining analyze network data, most models of networks are straightforward-focusing on many connections of a single type. In order to better understand relationships between different types of entities and extract meaningful structure from heterogeneous data, data mining algorithms need to be developed for new models of complex graphs. In this extended abstract, we describe some existing graph models and propose directions for more robust models that can serve as a backbone for analysis and mining of heterogeneous network data sets. We also identify possible metrics that attempt to capture the relationship of heterogeneous components of the network and serve to give insight into the topological relationships that exist among different node and edge types. The development of these different metrics will be important for designing meaningful clustering and graph mining algorithms for these data. Finally, we consider dynamic versions of these networks and present issues for the next generation of mining algorithms related to community identification, pruning and large graph approximations, privacy preservation of complex graph data.