Using Graph Algorithms to Detect Supernodes

Neo4j Graph Platform Graph Algorithms/Graph Data Science
1

Hi there,

I am trying to establish a process to detect supernodes using one of the graph algorithms. I'm currently using Degree Centrality, and am running into performance issues. I'm working with a very large graph >2 billion nodes, and when testing the degree centrality algorithm on a node of degree ~500, it's taking about 3 minutes to run. Of course, it won't be possible to get a degree for each node if that kind of performance is the norm. The whole purpose of this exercise is to detect supernodes, so once I get to one, if the degree centrality can't make the calculation, this approach isn't going to work.

Additionally, I am looking at using APOC to batch out the process.

Is anyone else trying to solve this problem? Any suggestions on how to approach it are most welcome.

Thanks!
Evan

2

Hi,

There are already some APOC routines that can help you with this.
https://neo4j-contrib.github.io/neo4j-apoc-procedures/#node-functions

You could use apoc.periodic.iterator to run several in parallel.

David

3

In case anyone stumbles on this, I found this query to be helpful:

MATCH (n)
WHERE apoc.nodes.isDense(n)
WITH labels(n) AS labelsList
UNWIND labelsList AS label
RETURN label, COUNT(*) AS denseNodesCount
ORDER BY denseNodesCount DESC
2 Likes