KAIST AI 대학원 신기정 교수님의 수업인 데이터마이닝 및 소셜 네트워크 분석 수업 필기입니다.
We have now seen that densification power laws and shrinking effective diameters are properties that hold across a range of diverse networks.
"How do we build models of network generations of evolution?"
Lattice Networks
Lattice networks have High clustering coefficients, Large diameter
Small-world Models
- Small-world Model have High clustering Low diameter, Huge giant connected component.
- Does not lead to the correct degree distribution, no heavy-tail
- Start with a low-dimensional regular lattice
Has high clustering coefficient - Rewire: introduce randomness for 'shortcuts'
For each edge with prob. p move the other end to a random node
Preferential Attachment
- Repeats adding a new node, create \(m\) out-links
- Probability of linking a node \(i\) is proportional to its degree \(d_i\)
$$ P = {d_i \over \sum d_j} $$
- Richer gets richer model(high degree일수록 새로운 node가 붙을 확률 증가)
- Leads to power-law in-degree distributions
- But all nodes have equal out-degree, does not have communities
$$ P(k) \propto k^{-3} $$
Edge Copying Model
- Power-law degree distributions
- Generates communities
- But, the diameter does not shrink
- Add a node and choose \(k\) the number of edges to add
- With prob \(\beta\) select \(k\) random vertices and link to them
- With prob \(1-\beta\), \(k\) edges are copied from a randomly chosen node
'GraphMining' 카테고리의 다른 글
Kronecker Graph Model (0) | 2022.09.26 |
---|---|
Community Guided Attachment & Forest Fire Model (0) | 2022.09.26 |
Structural properties (2) | 2022.09.14 |
Random Networks (0) | 2022.09.14 |
Intro & Basic Concepts (0) | 2022.09.06 |