Autoencode Your Networks
Post · Apr 12, 2017 03:14 ·
Autoencoding for networks. Struc2Vec gives you a new way to view and understand your networks.
Highlights From the Paper
- Although identification of such functions often leverage node and edge attributes, a more challenging and interesting scenario emerges when node function is defined solely by the network structure. In this context, not even the labels of the nodes matter but just their relationship to other nodes (edges).
- Two nodes that are structurally similar will be considered so, independently of their position in the network and node labels in their vicinity.
- On the one hand, structural identity is a concept independent of network position, while on the other hand, homophily is a concept tied to network proximity.
- struc2vec assesses the structural similarity of node pairs without leveraging node or edge attributes, including node labels.
- Daniel R. Figueiredo
- Leonardo F. R. Ribeiro
- Pedro H. P. Saverese
Structural identity is a concept of symmetry in which network nodes are identified according to the network structure and their relationship to other nodes. Structural identity has been studied in theory and practice over the past decades, but has only recently been addressed with techniques from representational learning. This work presents struc2vec, a novel and flexible framework for learning latent representations of node's structural identity. struc2vec assesses structural similarity without using node or edge attributes, uses a hierarchy to measure similarity at different scales, and constructs a multilayer graph to encode the structural similarities and generate structural context for nodes. Numerical experiments indicate that state-of-the-art techniques for learning node representations fail in capturing stronger notions of structural identity, while struc2vec exhibits much superior performance in this task, as it overcomes limitations of prior techniques.
Read the paper (pdf) »