Networks

Structure, centrality, graph models, and spreading on real datasets

Topology Changes the Dynamics

In network models, the interaction pattern is part of the system itself. Who can contact whom changes the outcome as much as the local update rule.

Session Arc

Start from data

Load a real graph, inspect its size, and understand what the nodes and edges mean.

Measure structure

Use centrality, clustering, components, and path lengths to characterize the network.

Build baselines

Compare the real graph against random, small-world, and scale-free models.

Add dynamics

Run SIS and SIR on top of the same topology and explain what changed.

Two Real Datasets Anchor the Session

Enron email

Communication graph between email addresses. Good for centrality, communities, robustness, and person-to-person spreading.

USA Flights

Mobility graph between airports built from OpenFlights data. Good for connectivity, hubs, and transport-style spreading.

NetworkX Workflow

Load

Read an edge list or GraphML file into networkx.

Inspect

Count nodes and edges, test connectivity, and look at node attributes.

Measure

Compute centrality, clustering, components, and path-based metrics.

Simulate

Run SIS or SIR directly on the graph and compare outcomes across structures.

Questions for Any Network

Local importance

Degree, closeness, betweenness, and PageRank tell different stories about node influence.

Global structure

Density, connected components, path lengths, and clustering quantify the overall shape of the graph.

Model comparison

Random, small-world, and scale-free graphs produce different path and hub structures.

Epidemic consequences

Hubs and bottlenecks strongly affect thresholds, outbreak size, and control strategies.

Standard Graph Models

Model Intuition What to inspect
Erdős-Rényi \(G(N, p)\) random edges baseline connectivity and path lengths
Watts-Strogatz \(WS(N, k, p)\) local links plus rewiring clustering and small-world effect
Barabási-Albert \(BA(N, m)\) preferential attachment hubs and heavy-tailed degree distributions

Spreading Rules on Graphs

SIS model

\[ \mathbb P(S \to I) = 1 - (1 - \beta)^m, \qquad \mathbb P(I \to S) = \mu \]

\(m\) is the number of infected neighbors.

SIR model

Use the same infection mechanism, but replace recovery back to susceptible with a permanent recovered state.

The Assignment Has Two Layers

Structure first

Characterize one real network with metrics, plots, and centrality rankings.

Dynamics second

Run both SIS and SIR on that same network and explain how the structure changes the propagation outcome.

Full Module Pages