· neo4j cypher

neo4j/cypher: Properties or relationships? It's easy to switch

I’ve written previously about how I’ve converted properties on nodes into relationships and over the past week there was an interesting discussion on the neo4j mailing list about where each is appropriate.

Jim gives quite a neat summary of the difference between the two on the thread:

Properties are the data that an entity like a node [or relationship] holds. Relationships simply form the semantic glue (type, direction, cardinality) between nodes.

To add to that, if you find yourself using a property to help narrow down a traversal then it’s worth checking whether you’d be better off modelling it as a relationship.

The neat thing is that it’s reasonably easy to change your mind at a later stage and convert a property into a relationship or vice versa.

For example let’s say we’re modelling some products for a shop. We might start with the following cypher code to create them:

CREATE (frenchConnection { name: "French Connection" })
CREATE  (dress1 { name: "Halter Dress", colour: "Blue"})-[:BRAND]-(frenchConnection)
CREATE  (dress2 { name: "Another Dress", colour: "Yellow"})-[:BRAND]-(frenchConnection)
CREATE  (dress3 { name: "Different Dress", colour: "Blue"})-[:BRAND]-(frenchConnection)
RETURN dress1, dress2, dress3
==> +--------------------------------------------------------------------------------------------------------------------------------------------+
==> | dress1                                     | dress2                                        | dress3                                        |
==> +--------------------------------------------------------------------------------------------------------------------------------------------+
==> | Node[2]{name:"Halter Dress",colour:"Blue"} | Node[3]{name:"Another Dress",colour:"Yellow"} | Node[4]{name:"Different Dress",colour:"Blue"} |
==> +--------------------------------------------------------------------------------------------------------------------------------------------+
==> 1 row
==> Nodes created: 4
==> Relationships created: 3
==> Properties set: 7
==> 179 ms

If we want to find all the French connection clothing which is blue we could write the following query:

START brand = node:node_auto_index(name="French Connection")
MATCH brand<-[:BRAND]-product
WHERE product.colour = "Blue"
RETURN brand, product
==> +-----------------------------------------------------------------------------------+
==> | brand                             | product                                       |
==> +-----------------------------------------------------------------------------------+
==> | Node[1]{name:"French Connection"} | Node[2]{name:"Halter Dress",colour:"Blue"}    |
==> | Node[1]{name:"French Connection"} | Node[3]{name:"Another Dress",colour:"Yellow"} |
==> | Node[1]{name:"French Connection"} | Node[4]{name:"Different Dress",colour:"Blue"} |
==> +-----------------------------------------------------------------------------------+
==> 3 rows
==> 69 ms

Since our query doesn’t return many records it’s not a problem that we have to look up the 'colour' property for every product but with a bigger data set the query would slow down considerably. We might therefore decide to create a 'COLOUR' relationship.

One way to do that would be by running the same cypher query as before except we also create a relationship to a colour node:

CREATE (blue { name: "Blue"})
START brand = node:node_auto_index(name="French Connection"),
      blue = node:node_auto_index(name="Blue")
MATCH brand<-[:BRAND]-product
WHERE product.colour = "Blue"

CREATE UNIQUE product-[:COLOUR]->blue
RETURN brand, product

Our query to find the blue French Connection products would now look like this:

START brand = node:node_auto_index(name="French Connection"),
      blue = node:node_auto_index(name="Blue")
MATCH brand<-[:BRAND]-product-[:COLOUR]->blue
RETURN brand, product
==> +-----------------------------------------------------------------------------------+
==> | brand                             | product                                       |
==> +-----------------------------------------------------------------------------------+
==> | Node[1]{name:"French Connection"} | Node[2]{name:"Halter Dress",colour:"Blue"}    |
==> | Node[1]{name:"French Connection"} | Node[4]{name:"Different Dress",colour:"Blue"} |
==> +-----------------------------------------------------------------------------------+
==> 2 rows
==> 3 ms

We can do it the other way around as well i.e. we can search for relationships and then set properties on nodes.

For example let’s say we want to mark the blue French connection clothing as being on sale but we weren’t interested in searching specifically for on sale items:

START brand = node:node_auto_index(name="French Connection"),
      blue = node:node_auto_index(name="Blue")
MATCH brand<-[:BRAND]-product-[:COLOUR]->blue
SET product.state = "OnSale"
RETURN brand, product

If we now repeat our previous query we’ll see that property on the products:

START brand = node:node_auto_index(name="French Connection"),
      blue = node:node_auto_index(name="Blue")
MATCH brand<-[:BRAND]-product-[:COLOUR]->blue
RETURN brand, product
==> +--------------------------------------------------------------------------------------------------+
==> | brand                             | product                                                      |
==> +--------------------------------------------------------------------------------------------------+
==> | Node[1]{name:"French Connection"} | Node[2]{name:"Halter Dress",colour:"Blue",state:"OnSale"}    |
==> | Node[1]{name:"French Connection"} | Node[4]{name:"Different Dress",colour:"Blue",state:"OnSale"} |
==> +--------------------------------------------------------------------------------------------------+
==> 2 rows
==> 2 ms

All the queries used in this post are on available as a gist if you’re interesting in playing around with it further. You can read more about cypher write syntax on the documentation pages.

  • LinkedIn
  • Tumblr
  • Reddit
  • Google+
  • Pinterest
  • Pocket