Procedures And Algorithms Overview

Consumer View

This is the part of Nexyron where the graph stops being only a place to store connected records and starts becoming an execution surface for retrieval, ranking, grouping, projection, and analysis.

The main practical categories are:

• catalog and projection procedures
• semantic and retrieval procedures
• graph algorithms

The catalog/projection family now also includes materialized aggregate-view lifecycle procedures, which means analytical summaries can be managed as first-class Cypher objects rather than only ad hoc query outputs.

Conceptual Explanation

Procedures matter because they let Nexyron extend normal Cypher without inventing a separate analytics DSL.

The pattern is:

1. CALL a procedure
2. YIELD rows
3. feed those rows back into MATCH, WITH, WHERE, RETURN, and functions

Simple example:

CALL nexyron.pagerank() YIELD node_id, score
MATCH (page)
WHERE id(page) = node_id
RETURN page.title AS page, score
ORDER BY score DESC
LIMIT 10

That lets users do things like:

• run PageRank on a named graph and then join scores back to nodes
• run graph_quality once to get cleanup, grouping, centrality, and weak-connection diagnostics for the observed or knowledge graph
• perform semantic retrieval and then filter by graph structure or time
• detect communities and then profile them with document metadata

Procedure Families

• projection and catalog procedures manage named analytical graphs and supporting metadata
• semantic procedures expose retrieval and ranking workflows that use embeddings, vector indexes, or diversity-aware ranking
• algorithm procedures expose centrality, traversal, similarity, community detection, flow, structure, and routing workloads

Community Detection Is The Most Misunderstood Family

This repo now has separate pages because “community detection” is not one feature.

• louvain is a fast modularity-oriented baseline
• leiden is usually the better production default when partition quality matters
• hierarchical_leiden preserves several resolution levels
• label_propagation is the quick heuristic option
• slpa, sllpa, and bigclam accept overlap instead of forcing one label
• infomap is about flow and path compression
• graph_quality is a bundled diagnostic procedure that computes cleanup, community, centrality, weak-node, and runtime metadata over one projected graph
• stochastic block procedures are about structural roles and interaction patterns
• modularity_metric and conductance score communities rather than generating them
• k1coloring and maxkcut are partitioning tools with goals that differ from community discovery

Recommended Reading Path

1. read procedures/README.md for the full family map
2. read the relevant family page such as community detection
3. read the per-procedure page for the exact behavior, selection guidance, use cases, and limitations