图机器学习基础知识——CS224W（14-traditional-generation）

CS224W: Machine Learning with Graphs

Stanford / Winter 2021

14-traditional-generation

Properties of Real-world Graphs

Properties of Real-world Graphs（第二章的回顾）

Degree Distribution

Degree Distribution

• Degree distribution $P(k)$: Probability that a randomly chosen node has degree $k$

  $$P(k)=N_k/N$$
  $N_k$: the number of nodes with degree $k$

Clustering Coefficient

Clustering Coefficient

• For one node

  $$C_i=\frac{2e_i}{k_i\left(k_i-1\right)}$$
  $e_i$ is the number of edges between the neighbors of node $i$; $k_i$: the degree of node $i$

• Graph clustering coefficient

  $$C=\frac{1}{N}\sum _i^N{C}_i$$

Connectivity

Connectivity

• Size of the largest connected component

• Largest component = Giant component

Path Length

Path Length

• Diameter: The maximum (shortest path) distance between any pair of nodes in a graph (任意两节点间的最大的最短路径)

• Average path length for a connected graph or a strongly connected directed graph

  $$\bar{h}=\frac{1}{2E_{\max }}\sum _{i,j\ne i}{h}_{ij}$$
  $h_{ij}$: the distance from node $i$ to node $j$; $E_{max}$: the max number of edges $n(n-1)/2$

All these models have prior assumption of the graph generation processes (以下传统图生成模型都有先验的假设)

Erdös-Renyi Random Graphs

Erdös-Renyi Random Graphs

• Two variants

  • $G_{np}$: undirected graph on $n$ nodes where each edge $(u,v)$ appears i.i.d. with probability $p$

  • $G_{nm}$: undirected graph with $n$ nodes, and $m$ edges picked uniformly at random

• Degree Distribution of $G_{np}$

  • Degree distribution of $G_{np}$ is binomial

  $$P(k)=\left(\begin{array}{c}n-1\\ k\end{array}\right)p^k(1-p{)}^{n-1-k}$$

• Clustering Coefficient of $G_{np}$

  • Expected $E[e_i]$

    $$p\frac{k_i\left(k_i-1\right)}{2}$$

  • $E[C_i]$

    $$\frac{p\cdot k_i\left(k_i-1\right)}{k_i\left(k_i-1\right)}=p=\frac{\bar{k}}{n-1}\approx \frac{\bar{k}}{n}$$

• Connected Components of $G_{np}$

  

• Def: Expansion

  

  

  

The Small-World Model

Paper : Collective dynamics of ‘small-world’ networks

The Small-World Model

• Key Idea: Interpolate between regular lattice graphs and $G_{np}$ random graph (同时满足高聚类系数以及低图直径)

  

• Small-World Model

  • Start with a low-dimensional regular lattice

    • In our case we are using a ring as a lattice

    • Has high clustering coefficient

    

  • Rewire: Introduce randomness (shortcuts)

    • Add/remove edges to create shortcuts to join remote parts of the lattice (随机创建shortcuts)

    • For each edge, with prob. $p$, move the other endpoint to a random node (对于每条边，以概率$p$随机移除其终点到一个随机点)

    

  

  

Kronecker Graph Model

Paper : Kronecker Graphs: An Approach to Modeling Networks

Kronecker Graph Model

• Key Idea: A recursive model of network structure

  

• Kronecker product

  Define a Kronecker product of two graphs as a Kronecker product of their adjacency matrices (邻接矩阵不断kronecker product)

  

  • Kronecker graph is obtrained by growing sequence of graphs by iterating the Kronecker product over the initiator matrix $K_1$

  $$K_1^{[\mathrm{m}]}=K_{\mathrm{m}}=\underset{\mathrm{m}\text{ times }}{\underbrace{K_1\otimes K_1\otimes \dots K_1}}=K_{\mathrm{m}-1}\otimes K_1$$

  

Stochastic Kronecker Graphs

Stochastic Kronecker Graphs

• Algorithm

  • Create $N_1\times N_1$ probability matrix ${\Theta }_1$ (为了引入随机性，将邻接矩阵改为概率矩阵)

  • Compute the $k^{th}$ Kronecker power ${\Theta }_k$

  • For each entry $p_{uv}$ of ${\Theta }_k$ include an edge $(u,v)$ in $K_k$ with probability $p_{uv}$

  

• Generation of Kronecker Graphs

  • 由于概率矩阵有$n^2$个元素，共需要翻转硬币$n^2$次，速度太慢

  • 根据Kronecker Product的性质，采取如下方式确定一条连边

    

    • 每一步都从4个大格子中根据概率选择一个进行深入，不断迭代选择直到最后一个特征元素，连边

    • 若两次连边冲突，忽略即可