cs229

huyi / September 2022

@import “https://haogeshuohuanihaohaoting.github.io/static/mdCreateMenu.js”

lecture 2: traditional feature-based method

overview

traditional ML pipeline

• Design features for nodes/links/graphs 设计node/link/graph level的特征
• Obtain features for all training data 获得这些特征

  这个 lecture 讲什么：

  传统的 ML 用的都是一些 handcrafted feature，这节课就讲这些features。

• 注意这些feature的类型和任务类型是对应的
  • node feature - node prediction
  • link feature - link prediction
  • graph feature - graph prediction

2.1 NODE -level tasks and features

2.1.1 overview

Goal: Characterize the structure and position of a node in the network:

• Node degree
• Node centrality
• Clustering coefficient
• Graphlets

  2.1.2 node degree $d_v$

• 有多少条边（邻居节点）。

• neighbour 之间不做区分

2.1.3 node centrality $c_v$

• centrality $c_v$ 同时把 node importance 考虑进来
• Different ways to model importance:
  • Engienvector centrality
  • Betweenness centrality
  • Closeness centrality
  • and many others…

    node centrality (1)

    

LINK

GRAPH