Analyzing Centralities of Embedded Nodes

slides for spotlight talk at ICDM workshop on Large Scale Graph Representation Learning and Applications.

Transcript

1. Analyzing Centralities of Embedded Nodes Kento Nozawa (AIST & University of Tokyo) Masanari Kimura (University of Tsukuba) Atsunori Kanemura (AIST & LeapMind Inc.) Nov. 17, 2018 @ GRLA2018 E-mail: k_nzw@klis.tsukuba.ac.jp Code: https://github.com/nzw0301/grla2018
2. K. Nozawa et al., Analyzing Centralities of Embedded Nodes. In GRLA. https://bit.ly/2PnUzgX, Nov. 17, 2018. Node embeddings Node embeddings are used as feature vectors of machine learning tasks. u = 0.1 ⋮ 0.7 Embedding ML task Where are misclassified nodes from in ML task? Classifier Node vector Graph
3. K. Nozawa et al., Analyzing Centralities of Embedded Nodes. In GRLA. https://bit.ly/2PnUzgX, Nov. 17, 2018. Analysis of the distributions of misclassified node centralities Misclassified nodes tend to have lower-centralities.
4. K. Nozawa et al., Analyzing Centralities of Embedded Nodes. In GRLA. https://bit.ly/2PnUzgX, Nov. 17, 2018. Conclusions • We analyze the distributions of misclassified node centralities. • Misclassified nodes tend to have lower centralities. • Future work: Developing a novel node embedding algorithm based on our analysis.