Research notes: network analysis

This Advanced note supports Network analysis in notebooks. For a first mixed-methods assignment, start with Your first research project.

Define the network before measuring it

Specify the unit represented by each node, the meaning and direction of each relationship, the collection period, and the inclusion rule. Distinguish an absent recorded relationship from evidence that no relationship exists. Record whether repeated edges are meaningful or duplicate imports.

The worked guide uses a synthetic graph and the native construction API. graphforge.datasets is a proposed catalogue, not an installed loader. For a real dataset, retain its exact source, license, preparation code, and schema; use graph construction for import.

Interpret results within their scope

• Degree counts depend on whether you count edges or distinct neighbors and whether incoming, outgoing, or undirected relationships are included.
• Use count(relationship) after an OPTIONAL MATCH to preserve zero-degree nodes without counting the unmatched row as an edge.
• Community labels describe an algorithmic partition. They do not identify social groups, interview themes, or shared beliefs without further evidence.
• A path demonstrates recorded connectivity, not causation or transmission.
• Centrality does not by itself measure expertise, credibility, or influence.

Connect a numerical pattern back to source material and inspect exceptions. Do not infer a population relationship from a convenience sample merely because its graph query has a deterministic answer.

Reproduce the analysis

Record selected labels, relationship types, direction, algorithm options, and source scope. Public result identities are UUIDs; do not depend on internal numeric storage IDs. Python results are Arrow tables. See analytics integration for explicit pandas/NetworkX/igraph conversion, and the algorithm catalog for native verbs.

Use save and reopen for a durable working graph and portable projects to share it. A saved score is an observation about the graph used to compute it, not an automatically updated property of later graph states.