SET in combination with CASE statement in cypher

Question

I am tryin to set two different relationship properties to a count, with a case construct depending on the value of another relationship property. There is a console at http://console.neo4j.org/?id=rt1ld5

the cnt column contains the number of times r.value occurs. The two first rows of the initial query in the console indicate that the term "Car" is linked to 1 document that is considered relevant, and to two documents that are considered not relevant.

I want to SET a property on the [:INTEREST] relation between (user) and (term) with two properties, indicating how many times an interest is linked to a document that is considered relevant or not. So for (John)-[r:INTEREST]->(Car) I want r.poscnt=1 and r.negcnt=2

I.m struggling with the CASE construct. I tried various ways, this was the closest I got.

MATCH (u:user)-[int:INTEREST]->(t:term)<-[:ISABOUT]-(d:doc)<-    [r:RELEVANCE]-(u)
WITH int, t.name, r.value, count(*) AS cnt
CASE
  WHEN r.value=1 THEN SET int.poscnt=cnt
  WHEN r.value=-1 THEN SET int.negcnt=cnt
END

But it's returning an error

Error: Invalid input 'A': expected 'r/R' (line 3, column 2)
"CASE"
  ^

Answer 1

This did it! Also see console at http://console.neo4j.org/?id=rq2i7j

MATCH (u:user)-[int:INTEREST]->(t:term)<-[:ISABOUT]-(d:doc)<-[r:RELEVANCE]-(u)
WITH int, t, 
     SUM(CASE WHEN r.value= 1 THEN 1 ELSE 0 END ) AS poscnt, 
     SUM(CASE WHEN r.value= -1 THEN 1 ELSE 0 END ) AS negcnt
SET int.pos=poscnt,int.neg=negcnt
RETURN t.name,int.pos,int.neg

Answer 2

Is it important for you to keep positive and negative count separate? It seems you could have a score property summing positive and negative values.

MATCH (u:user)-[int:INTEREST]->()<-[:ISABOUT]-()<-[r:RELEVANCE]-(u)
SET int.score = SUM(r.value)
RETURN t.name, int.score

You already seem to have found a working solution but I'll add a note about CASE as I understand it. While CASE provides branching, I think it's correct to say that it is an expression and not a statement. It resembles a ternary operator more than a conditional statement. As the expression

a > b ? x : y;

is resolved to a value, either x or y, that can be used in a statement, so also

CASE WHEN a > b THEN x ELSE y END

resolves to a value. You can then assign this value

result = CASE WHEN a > b THEN x ELSE y END

Your original query used the CASE expression like a conditional statement

CASE WHEN a > b THEN result = x ELSE result = y END

which resembles if-else

if a > b { result = x; } else { result = y; }

Someone may want to correct the terminology, the point is that in your working query you correctly let CASE resolve to a value to be used by SUM rather than put a conditional assignment inside CASE.