科研 Skill

社会网络结构分析

alterlab-sna

Computes network centrality, community detection, and ERGM inferential models matched to the research question via networkx/igraph. Use for social-network analysis beyond basic visualization.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

Social Network Analysis — Match the Measure to the Question

Skill type: ANALYSIS MODULE. SNA treats the ties as the data. The discipline is not running a centrality function — it is choosing the centrality that answers the substantive question, respecting network dependence (ties are not independent observations), and using an inferential model (ERGM) when you want to explain tie formation rather than describe it. Computation is handed to alterlab-networkx.

Core Mission

THE TIE IS THE UNIT. PICK THE CENTRALITY THAT MATCHES THE QUESTION; TIES ARE NOT IID.

When to Use This Skill

  • "Run a social network analysis / compute centrality on my network."
  • "Who are the key players / brokers in this network?"
  • "Detect communities / clusters in my network."
  • "Fit an ERGM to explain why ties form."

Does NOT Trigger

The request is really about…Route toWhy not this skill
General graph algorithms / plotting / I/Oalterlab-networkxThis skill adds SNA method choice; networkx executes.
Graph neural networks / node embeddings for predictionalterlab-torch-geometricDeep learning on graphs, not SNA description/inference.
Whether a relational design is appropriatealterlab-ssci-design-gateDesign routing, upstream.
Plain tabular statisticsalterlab-statistical-analysisNo network structure.

Hand-off rule. When the request matches a row above, open your answer by naming that skill by its exact id (for example alterlab-networkx) and saying in one line why it fits better, then stop applying this skill's procedure. Name it even when you also need to ask for missing input. If only part of the request is out of scope, name the skill for that part and do only the in-scope part. A brief general pointer is fine; the named skill does the work.

Running the code. In Claude Code, run analysis code in a throwaway environment instead of installing packages: uv run --python ">=3.11" --with "networkx>=3.4" --with pandas python network.py. This needs no install approval and leaves the user's environment untouched. If uv or the network is unavailable, say so and hand over the script; never report results you did not compute.

Centrality — pick by the substantive claim (verified networkx 3.x)

QuestionCentralityCall
Who has the most direct ties?degreenx.degree_centrality(G)
Who bridges otherwise-separate groups?betweenness (brokerage)nx.betweenness_centrality(G)
Who can reach everyone fastest?closenessnx.closeness_centrality(G)
Who is connected to well-connected others?eigenvector / PageRanknx.eigenvector_centrality(G) / nx.pagerank(G)

Do not report "centrality" generically — name which one and why it matches the claim (a broker argument needs betweenness, not degree).

Community detection (verified)

  • nx.community.louvain_communities(G) — native in networkx 3.x (no python-louvain needed).
  • nx.community.greedy_modularity_communities(G) — Clauset-Newman-Moore.
  • nx.community.modularity(G, communities) — score a partition.
  • Leiden (Traag et al.) guarantees well-connected communities and is the recommended improvement over Louvain. networkx ≥ 3.7 ships it natively: nx.community.leiden_communities(G, metric="modularity", seed=...) (on 3.5–3.6 the function exists but is backend-only). For large graphs use igraph g.community_leiden(objective_function="modularity") or leidenalg. Both networkx and igraph default to the CPM objective, which at resolution 1 can return near-singleton partitions — pass the modularity objective (or a deliberate CPM resolution) explicitly.

Inference: ERGM (no mature Python — use R statnet)

To explain tie formation (reciprocity, homophily, triadic closure) rather than describe it, use an exponential random graph model. There is no mature Python ERGM package (pyERGM exists but is young); the standard is R statnet/ergm via an R bridge:

library(ergm)
model <- ergm(net ~ edges + mutual + nodematch("group") + gwesp(0.5, fixed = TRUE))
summary(model)

Declare the R dependency; do not fabricate a Python ERGM API. Full algorithm-selection guidance, the ERGM term glossary, and the dependence caveat: references/sna_methods.md.

The dependence caveat

Network ties violate the independence assumption of ordinary regression: you cannot regress an outcome on centrality and read the SE naively, because nodes are linked. Use permutation/QAP tests for dyadic hypotheses, or ERGM/latent-space models for tie-level inference. Flag this whenever a network measure is fed into a downstream test.

Output Template

NETWORK:      <directed? weighted? n nodes, m edges; how ties were defined>
MEASURE:      <named centrality/community algorithm + why it matches the question>
RESULT:       <top nodes / partition + modularity; not just a number>
INFERENCE:    <ERGM terms + fit, OR permutation/QAP test> — respect tie dependence
CLAIM SCOPE:  descriptive (structure) vs inferential (tie-formation) — do not conflate

References

  • references/sna_methods.md — centrality selection, Louvain vs Leiden, ERGM terms + R bridge, QAP/permutation, large-graph tooling.

Part of the AlterLab Academic Skills suite.

示例提示词

帮我计算这份关系数据的中心性指标,并做社区发现分析。

所属插件

社会学、政治学与城市研究面向社会科学研究:问卷与访谈设计、质性编码、文本即数据、社会网络、ABM、比较政治与选举数据、RCT 与 DID、城市与区域分析。

安装该插件会一并获得同领域的 58 个 Skill / MCP。

同插件的其他科研 Skill

常见问题

社会网络结构分析能做什么?

Computes network centrality, community detection, and ERGM inferential models matched to the research question via networkx/igraph. Use for social-network analysis beyond basic visualization。

如何安装社会网络结构分析?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「社会网络结构分析」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

社会网络结构分析需要哪些权限?

该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。

社会网络结构分析经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。