Reading networks – content extraction from complex networks
This event is part of the Biophysics/Condensed Matter Seminar Series.
Many real world networks contain signatures of the nodes function. However, the same networks contain a myriad of extra edges, which hide this inherent signature. I will discuss the definition of such a signature and propose multiple supervised and unsupervised methods to detect nodes with similar function. The most prevalent unsupervised approach for node classification relies on decomposition of a network into communities. A fundamental assumption underlies community detection is that nodes of identical function are located in the same dense sub-network. However, nodes with similar classifications may have similar connection patterns in the network, even if they reside in remote regions of the network. We introduce a novel method for the detection of groups of non-adjacent nodes with similar function in networks through the similarity of measures on the network surrounding them. When tested in four real world networks with ground truth classifications, the groups detected by our algorithm were significantly more homogenous than those found by common community detection algorithms. When used in a supervised context, precise predictions of vertices function can be accomplished.