Skip to content
GraphForge v0.6.0

Network analysis in notebooks

Advanced: basic Python, database queries, and terminal use.

This optional guide is for readers who can run Python and want to inspect a network. Start with a notebook and your first graph. For a first social-science project combining numerical observations and interpreted passages, use Your first research project.

The examples use a synthetic citation network. Its counts describe the entered relationships; they do not measure a paper’s quality, prove influence, or estimate a population characteristic. Install the matching version through Installation.

Create a small network

Section titled “Create a small network”

Run the Python blocks in order in one session. No dataset download is needed. GraphForge does not ship the proposed graphforge.datasets catalogue. For real data, obtain and document your source, then use the supported construction APIs.

from graphforge import GraphForge


forge = GraphForge()
survey = forge.add_node("Paper", title="Survey")
methods = forge.add_node("Paper", title="Methods")
replication = forge.add_node("Paper", title="Replication")
forge.add_edge(survey, "CITES", methods)
forge.add_edge(survey, "CITES", replication)
forge.add_edge(replication, "CITES", methods)

One node represents a paper; a directed CITES relationship runs from the citing paper to the cited paper. Before analyzing another network, state what its nodes and relationships represent, whether direction matters, and which records are missing. Two datasets using the word “connection” need not measure the same thing.

Degree distribution

Section titled “Degree distribution”

Count incoming citations for every paper, including papers with no citations:

degrees = forge.execute("""
    MATCH (paper:Paper)
    OPTIONAL MATCH (:Paper)-[citation:CITES]->(paper)
    RETURN paper.title AS title, count(citation) AS citations
    ORDER BY citations DESC, title
""")
print(degrees.to_pylist())

Expected output:

[{'title': 'Methods', 'citations': 2}, {'title': 'Replication', 'citations': 1}, {'title': 'Survey', 'citations': 0}]

count(citation) excludes the null relationship from an unmatched OPTIONAL MATCH. Using count(*) here would incorrectly count that row as one citation. Decide whether you need a relationship count or a count of distinct neighbors: repeated relationships can make them different quantities.

Path analysis

Section titled “Path analysis”

Inspect the direct and indirect routes from Survey to Methods:

paths = forge.execute("""
    MATCH path = (source:Paper {title: $source})-[:CITES*1..2]->(target:Paper {title: $target})
    RETURN length(path) AS hops
    ORDER BY hops
""", {"source": "Survey", "target": "Methods"})
print(paths.to_pylist())

Expected output: [{'hops': 1}, {'hops': 2}].

These are paths in the recorded citation graph, not proof that one author read another or that an idea spread along that route. Variable-length queries can enumerate many paths; choose a justified hop bound. ORDER BY ... LIMIT limits the returned rows and does not by itself establish a bound on traversal work. See native path algorithms for other tasks.

Graph algorithms with analyst verbs

Section titled “Graph algorithms with analyst verbs”
ranking = forge.rank("Paper", by="pagerank", via="CITES", directed=True)
print(ranking.select(["title", "score"]).to_pylist())

PageRank summarizes this graph’s structure. Scores depend on the graph, direction, and algorithm settings; a high score is not a measure of truth, research quality, or causal importance.

For a larger network where grouping is useful, community detection is optional:

communities = forge.cluster("Paper", by="louvain", via="CITES")
print(communities.select(["title", "community_id"]).to_pylist())

A computed community is an algorithmic grouping. Inspect its members and edges before assigning a social meaning or comparing it with interview themes. A three-paper teaching graph does not establish useful real-world communities. The algorithm catalog defines the available verbs and options. rank and cluster support explicit write_property; writing a score does not give it permanent validity after the graph changes.

Integration with pandas

Section titled “Integration with pandas”

Python queries and analyst verbs return PyArrow tables. Use to_pylist() for ordinary rows or degrees.to_pandas() if you installed pandas. There are no CypherValue wrappers or row .value calls. For plotting and explicit NetworkX/igraph construction from selected Arrow results, use analytics integration and visualization examples. Those integrations do not require an alternate graph execution engine inside GraphForge.

Persisting analysis results

Section titled “Persisting analysis results”
forge.close()

This example is memory-only. To keep a graph across kernel resets, follow save and reopen using an existing directory on supported storage. Keep original data and the code that constructs and analyzes it. Use portable export, verification, and import to transfer a project; do not copy live project storage or commit it to Git.

Next steps

Section titled “Next steps”