hierarchicalfacets

= Using hierarchical facets 

We have a usecase with documents that are tagged with keywords in a theasaurus. This gists explains the model and is at the same time an invitation to suggest improvements. Because it would be nice to have something that performs better.

== The model

//setup
//hide
[source,cypher]
----
CREATE (doc1:`doc` {`name`:"doc 1"})
CREATE (doc2:`doc` {`name`:"doc 2"})
CREATE (doc3:`doc` {`name`:"doc 3"})
CREATE (doc4:`doc` {`name`:"doc 4"})
CREATE (doc5:`doc` {`name`:"doc 5"})
CREATE (root:`term` {`name`:"root"})
CREATE (term2:`term` {`name`:"term 2"})
CREATE (term3:`term` {`name`:"term 3"})
CREATE (term4:`term` {`name`:"term 4"})
CREATE (term5:`term` {`name`:"term 5"})
CREATE (term6:`term` {`name`:"term 6"})
CREATE (term7:`term` {`name`:"term 7"})
CREATE (term2)-[:BT]->(root)
CREATE (term3)-[:BT]->(root)
CREATE (term4)-[:BT]->(term2)
CREATE (term5)-[:BT]->(term2)
CREATE (term6)-[:BT]->(term3)
CREATE (term7)-[:BT]->(term3)
CREATE (doc1)-[:HAS_TERM]->(term2)
CREATE (doc1)-[:HAS_TERM]->(term4)
CREATE (doc1)-[:HAS_TERM]->(term5)
CREATE (doc2)-[:HAS_TERM]->(root)
CREATE (doc2)-[:HAS_TERM]->(term3)
CREATE (doc3)-[:HAS_TERM]->(term6)
CREATE (doc3)-[:HAS_TERM]->(term4)
CREATE (doc3)-[:HAS_TERM]->(term2)
CREATE (doc4)-[:HAS_TERM]->(term7)
CREATE (doc5)-[:HAS_TERM]->(term2)
CREATE (doc5)-[:HAS_TERM]->(term3)
----

//graph

== Get docs for each term

=== DIRECT
This query returns the docs to which the specified term is directly  (the term has a link to the doc) 

[source,cypher]
----
MATCH (d:doc)-[:HAS_TERM]->(t:term)
RETURN t.name AS term,collect(d.name) AS docs 
ORDER BY term.name
----

//table

=== DIRECT and INDIRECT
This query returns the docs to which the specified term is linked, both directly  (the term has a link to the doc) and indirectly  (one of the more detailed terms is linked to the doc)

[source,cypher]
----
MATCH (d:doc)-[:HAS_TERM]->()-[:BT*0..]->(t)
RETURN t.name AS term,collect(d.name) AS docs 
ORDER BY term.name
----

//table

== Get count for root and its children

[source,cypher]
----
match (d:doc)-[:HAS_TERM]->()-[:BT*0..]->(t2:term {name:"root"}) 
return t.name AS term, count(d) as docs 
order by docs desc
UNION
match (d:doc)-[:HAS_TERM]->()-[:BT*0..]->(t:term)-[:BT]->(t2:term {name:"root"}) 
return t.name AS term, count(d) as docs 
order by docs desc
----

//table


== Get count for "term 3" and its children

[source,cypher]
----
match (d:doc)-[:HAS_TERM]->()-[:BT*0..]->(t2:term {name:"term 3"}) 
return t.name AS term, count(d) as docs 
order by docs desc
UNION
match (d:doc)-[:HAS_TERM]->()-[:BT*0..]->(t:term)-[:BT]->(t2:term {name:"term 3"}) 
return t.name AS term, count(d) as docs 
order by docs desc
----

//table

== Question

The main question is, is there a faster way based on pure Neo4j? Trying this approach with 10k docs, a thesausrus  with 12k terms and 1.2M [:HAS_TERM] relationships takes over 20 secs when getting counts for "root" and diirect children.