Communication Network Analysis
Overview
This Skill treats a network as a research design, not a visualization. It locks what
nodes and edges mean, how the graph was observed, and which relational claim the data
can support before selecting a metric or model.
Required start
- State the communication phenomenon, goal class, and target quantity.
- Write the graph contract in references/network-protocol.md.
- Distinguish description, prediction, diffusion, selection, influence, and causal
intervention; they require different evidence.
Method routing
- centrality and community detection describe a specified graph;
- ERGM models cross-sectional tie dependence, conditional on the specified terms;
- SAOM/SIENA models co-evolution of ties and actor attributes in panel networks;
- relational-event models use event order and timing rather than aggregated ties;
- cascade and diffusion models require explicit exposure and adoption definitions;
- temporal, multiplex, bipartite, and multilayer networks retain their native structure
unless a justified projection or aggregation is tested for sensitivity.
Validation
- test direction, weighting, duplicate ties, self-loops, isolates, components, and time
ordering on known cases;
- report missing nodes/edges, API and deletion coverage, boundary sensitivity, bot or
automation rules, and projection loss;
- compare plausible graph constructions before interpreting centrality or communities;
- separate homophily, shared context, and prior similarity from social influence.
Output shape
Return the graph contract, construction audit, selected method and rejected
alternatives, diagnostics, sensitivity plan, and a claim map stating what each network
quantity can and cannot mean.
Hard boundaries
- Centrality is not influence, brokerage, authority, or importance without a mechanism.
- Community detection does not discover natural social groups by itself.
- Contemporaneous similarity does not identify contagion.
- A platform crawl is not automatically the whole network.
- Layout coordinates are visual encodings, not analytical results.
1---2name: communication-network-analysis3description: Design and audit communication network, diffusion, cascade, community, ERGM, SAOM, and multiplex analyses with explicit graph boundaries and selection-versus-influence claims. Use for 传播网络/扩散/级联/社群/ERGM/SIENA/关系数据. Do not use for non-relational content analysis or spatial proximity without ties.4---56# Communication Network Analysis78## Overview910This Skill treats a network as a research design, not a visualization. It locks what11nodes and edges mean, how the graph was observed, and which relational claim the data12can support before selecting a metric or model.1314## Required start15161. State the communication phenomenon, goal class, and target quantity.172. Write the graph contract in [references/network-protocol.md](references/network-protocol.md).183. Distinguish description, prediction, diffusion, selection, influence, and causal19 intervention; they require different evidence.2021## Method routing2223- centrality and community detection describe a specified graph;24- ERGM models cross-sectional tie dependence, conditional on the specified terms;25- SAOM/SIENA models co-evolution of ties and actor attributes in panel networks;26- relational-event models use event order and timing rather than aggregated ties;27- cascade and diffusion models require explicit exposure and adoption definitions;28- temporal, multiplex, bipartite, and multilayer networks retain their native structure29 unless a justified projection or aggregation is tested for sensitivity.3031## Validation3233- test direction, weighting, duplicate ties, self-loops, isolates, components, and time34 ordering on known cases;35- report missing nodes/edges, API and deletion coverage, boundary sensitivity, bot or36 automation rules, and projection loss;37- compare plausible graph constructions before interpreting centrality or communities;38- separate homophily, shared context, and prior similarity from social influence.3940## Output shape4142Return the graph contract, construction audit, selected method and rejected43alternatives, diagnostics, sensitivity plan, and a claim map stating what each network44quantity can and cannot mean.4546## Hard boundaries4748- Centrality is not influence, brokerage, authority, or importance without a mechanism.49- Community detection does not discover natural social groups by itself.50- Contemporaneous similarity does not identify contagion.51- A platform crawl is not automatically the whole network.52- Layout coordinates are visual encodings, not analytical results.
haochengw372-hash