How to do 'keyword' based search on neo4j across different node types?

[lingvisa]

The typical search in neo4j is to write query with restrictions and the node type is an important restriction type. For example:

match (n:Product) where n.name='iPhone' return n

However, if the user doesn't know what node type should be given, he may just write query:
match (n) where n.name='iPhone' return n

And this 'iPhone' may be names of one or another node types, such as Product, or Brand. To make the query more efficient, I may have a function working behind the scene and convert user's keyword search into this query:

match (n)
where (n:Product or n:Brand) and n._name = 'iPhone' 
return n

This takes 150 ms shown in the browser, while this one only takes 1 ms:

match (n:Product) where n._name = 'iPhone' 
return n

Is there a more efficient way to support users' such queries? Given a keyword without other restrictions, the query return any nodes that may hit this query.

[Cobra]

Hello @lingvisa

Did you have a look at full-text search?

Regards,
Cobra

[lingvisa]

I know of full-text search. Can I build a single fulltext index from different nodes? I used to create fulltext index on a single property of a single node. In the use case above, how can I create a single index to meet the requirements?

[clem]

I don't know if you know this, but Nodes can have multiple Labels.

You could make a Label that is a Superset of all types of nodes you want to query, e.g. BrandProduct.

When you create a node, then you would have:

CREATE (p:Brand:BrandProduct {name:'iPhone'}) or CREATE (p:Product:BrandProduct {name:'iPhone'})

Then your query would be:

match (n:BrandProduct) where n._name = 'iPhone' 
return n

You can also add new Labels to existing Nodes.

[lingvisa]

I could think about this, by adding additional label to exsiting nodes.

[ameyasoft]

Try this:

MATCH (n) 
WHERE exists (n.name) and labels(n) in [['Product], ['Brand']
and n.name = 'iphone'
RETURN n

[lingvisa]

@ameyasoft It works, but it's a big slower than this version:

match (n) where n.name = 'iPhone
 return n

All takes over 100 ms. I already created index on the 'name' property for all node types. Why isn't faster?

Only this takes 1 ms:

match (n:Product) where n.name='iPhone'
return n

[clem]

You can Profile this (using Neo4J's PROFILE or EXPLAIN), but I suspect that in the first version, (without specifying the Label) means there are a lot more nodes to go through.

I believe by limiting the Nodes to one type ('Product'), there's less to go through.

[Cobra]

You only need one full-text index, you can create it like this:

CALL db.index.fulltext.createNodeIndex("nameProductBrand", ["Product", "Brand"], ["name"])

To use it:

CALL db.index.fulltext.queryNodes("nameProductBrand", "iPhone") YIELD node, score
RETURN node, score

It should be the fastest method

Regards,
Cobra

[lingvisa]

@Cobra I will try this, probably it's what I need.