FP7 Logo

EC Project 257859

European Union

Co-funded by the 
European Union
 

Using KONECT to Investigate Graph Sampling Techniques

The University of Koblenz-Landau is currently investigating graph sampling techniques for very large networks. The networks studied in the ROBUST projects such as the business communities of IBM and SAP have such a large size that most conventional network mining algorithms cannot be applied to them easily: One must use clusters of servers to process the whole networks, necessitating huge amounts of computing power, memory and network bandwidth.

As an alternative, the University of Koblenz-Landau is investigating Graph Sampling techniques.  Graph sampling techniques are a set of methods in which a network is "summarized" to a smaller network having the same statistical properties.  This allows complex network mining techniques to be applied to the small, sampled networks, making them scalable.  The current research consists in determining which sampling techniques result in the most faithful sampled graphs, in function of the chosen network mining technique.

Comments (0)