Neo4j create multiple nodes and relationships

5
votes

I am using Neo4j to build a huge graph database (over a million nodes). The way I'm doing it right now is running a cypher CREATE (n {property:'value'}) query for each of the node. As expected, this is quite an inefficient method and it takes lots of time. Can somebody suggest me some alternative method to overcome this issue? I've heard that Neo4j also provides some default batch interface to create multiple nodes. I'm currently using this version of code (including the relationship):

create (a { name: "a" })-[:rel1]->(b {name : "b"}),(c {name: "c"})-[:rel2]->(d {name:"d"}),...

Is it an efficient method or are there any better methods? Thanks in advance! :)

3
neo4jcypher

3 Answers

5
votes

Two alternatives I'd consider:

  1. Use the LOAD CSV support: http://docs.neo4j.org/chunked/stable/query-load-csv.html
  2. Use a single paramaterized Cypher statement, and supply an array of parameters: http://docs.neo4j.org/chunked/stable/query-create.html#create-create-multiple-nodes-with-a-parameter-for-their-properties

Otherwise there is the batch inserter, as mentioned.

2
votes

There is an overview on the Neo4j website: http://www.neo4j.org/develop/import

But to be short, for more than a million nodes I would say:

If you can code in Java:

http://docs.neo4j.org/chunked/stable/batchinsert.html

else:

https://github.com/jexp/batch-import

0
votes

You can use unwind to your advantage. Assuming you are using the Java API, here's a small snippet using Groovy

List<Map<String, Object>> props = yourList.collect {["prop1": it.prop1, "prop2":it.prop2]}
String q = "UNWIND {props} AS p create(a {name: p.prop1})-[:rel1]->(b {name: p.prop2}) "
neo4jOperations.query(q, ["props" : props])

I've had great success with this method. You should be able to see a speedup of 5-30 times depending on your batch size