[GNN] Node2Vec: Scalable Feature Learning for Networks

0. Abstract

• Previsous Study
  • predction task -> feature를 학습시키는 과정에서 많은 발전
  • BUT connectivity pattern을 제대로 학습 못함
• This Paper
  • low dimensional feature learning & maximizing likelihoods of nodes
  • flexible notion of nodes ("neighborhood node" 너무 엄격한 구분 X)

1. Node2Vec

• semi supervisded algorithm 
  • supervised : expensive for real-word
  • unsupervised : generalize X
• SGD motivation
• #Flexibility #Generalization
  • u, s6 : hub node -> real word : mixture of such equivalence

• individual node -> extended to pairs of node
• objective function

objective function

+ BFS & DFS

https://www.theteams.kr/teams/2849/post/73221

그래프 임베딩 요약 by 왓챠

• BFS : 너비 우선 탐색
  • 현재정점에서 연결된 가까운 노드부터
• DFS : 깊이 우선 탐색
  • 현재 정점에서 들어갈 수 있는 노드까지