Neo4J variable-length pattern

Question

Is it possible to write a Cypher query with a variable-length patter, that specifies that nodes between the start and end node can only be nodes with a certain property?

I think that a query like the following would match all of the below patterns:

MATCH p=(n:Node {foo:"True"})<-[:REL*0..2]-(:Node)<-[:REL]-(:Node {bar:"True"})

(n:Node {foo:"True"})                                        <-[:REL]-(:Node {bar:"True"})
(n:Node {foo:"True"})                    <-[:REL]-(:Node)    <-[:REL]-(:Node {bar:"True"})
(n:Node {foo:"True"})  <-[:REL]-(:Node)  <-[:REL]-(:Node)    <-[:REL]-(:Node {bar:"True"})

I imagine it "copying" the <-[:REL]-(:Node) part of the query 0 to 2 times in between the (n:Node {foo:"True"}) and <-[:REL]-(:Node {bar:"True"}) query parts and constructing some kind of UNION of the matches.

Is this the right way to think about it? And how would i make the variable-lenght query match only paths where the middle nodes have the property {bar:"True"} in the middle nodes like so:

(n:Node {foo:"True"})                                                    <-[:REL]-(:Node {bar:"True"})
(n:Node {foo:"True"})                          <-[:REL]-({bar:"True})    <-[:REL]-(:Node {bar:"True"})
(n:Node {foo:"True"})  <-[:REL]-({bar:"True})  <-[:REL]-({bar:"True})    <-[:REL]-(:Node {bar:"True"})

Would the query below be the right approach?

MATCH p=(n:Node {foo:"True"})<-[:REL*0..2]-(:Node {bar:"True"})<-[:REL]-(:Node {bar:"True"})

It appears to me that this is not the case. Can somebody clearify on where I am going wrong in my thinking process?

You can use APOC with nodes filter: neo4j.com/labs/apoc/4.0/graph-querying — nimrod serok
– nimrod serok, Commented Apr 7, 2022 at 15:29
It'd be easy if "bar" was a label or property of the relationship - if that's compatible with your data model. — David Pond
– David Pond, Commented Apr 7, 2022 at 16:25
@DavidPond unfortunately it is only a node property - like in my example given, but please feel free to provide a solution if the property were of the relationship (although for my case it is not usable) — DigitalJedi
– DigitalJedi, Commented Apr 7, 2022 at 16:27
@nimrodserok My question still remains...How can I think about the variable-length pattern. And is my assumption correct? — DigitalJedi
– DigitalJedi, Commented Apr 7, 2022 at 16:28

David Pond · Accepted Answer · 2022-04-08 08:21:48Z

1

MATCH path=(:Node {foo:True})-[:REL*0..2]-(n:Node {bar:True})
WHERE all(x in nodes(path) WHERE x.bar=True OR x.foo=True)
RETURN n.name

answered Apr 8, 2022 at 8:21

David Pond

1,1359 silver badges16 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

David Pond · Accepted Answer · 2022-04-08 07:28:12Z

0

I think it'd be easy if foo were a relationship property, instead of a node property.

Create a simple example:

MERGE (a:Node {name:"a", foo:True})
MERGE (b:Node {name:"b", bar:True})
MERGE (c:Node {name:"c", bar:True})
MERGE (d:Node {name:"d", bar:False})
MERGE (e:Node {name:"e", bar:True})
MERGE (a)<-[:REL]-(b)
MERGE (b)<-[:REL]-(c)
MERGE (a)<-[:REL]-(d)
MERGE (d)<-[:REL]-(e)

I think you want to match nodes b and c, but not e because we'd have to traverse d which has foo:False.

Make bar a property on the relationships:

MATCH ()<-[r:REL]-(n:Node)
SET r.bar = n.bar

Now you can write:

MATCH (:Node {foo:True})<-[:REL*0..2 {bar:True}]-(n:Node)
RETURN n.name

Which returns b and c as required.

answered Apr 8, 2022 at 7:28

David Pond

1,1359 silver badges16 bronze badges

1 Comment

DigitalJedi Over a year ago

Looks easy and definitely works! BUT my dataset is about 150 million nodes and I dont want to change the relationships this way.

Collectives™ on Stack Overflow

Neo4J variable-length pattern

2 Answers 2

Comments

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

Comments

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related