OrientDB - Compute User Similarity - orientdb

Question

I am trying to solve a simple problem: compute user to user similarity based on euclidean distance computed for ratings of a product.
I am using a such query
SELECT U1.UserId, U2.UserId
FROM (
MATCH
{class:User, as: U1, where: (UserId=12345) } -rate-> {class:Product, as:P},
{class:User, as: U2, where: (UserId<>12345)} -rate-> {as:OP},
RETURN U1, U2, P, OP
)
I would now compute for each pair (user U1, user U2) a result that represents the distance between rating for common product.
Example of common products for users
U1,Product,Rating
1, xxx, 5
2, xxx, 2
1, yyy, 10
2, yyy, 8
So I would compute Sqrt((5-2)^2 + (10-8)^2) as Distance
Is this possible with a single query on OrientDB. Neo4J provide the WITH statement to manipulate consecutive instance in Cypher Query.
Many thanks in advance for all the help you can provide us.
Thx
Roberto

Answer 1

First of all I'd rewrite the MATCH statement to return the distance of ratings for two users and a product:
MATCH
{class:User, as: U1, where: (UserId=12345) }.outE("rate"){as:r1}.inV(){class:Product, as:P},
{class:User, as: U2, where: (UserId<>12345)}.outE("rate"){as:r2}.inV(){as:P},
RETURN U1, U2, (r1.rating - r2.rating) * (r1.rating - r2.rating) as squareDistance, P
Then you can use some outer SELECT to do the calculations:
SELECT U1, U2, P, sqrt(squareSum) as distance from (
SELECT U1, U2, P, sum(squareDistance) as squareSum from (
MATCH...
) GROUP BY U1, U2, P
)
The only problem here is that OrientDB does not have a built-in sqrt() function, so you have to write your own sqrt() in javascript. It's quite easy, because inside js functions you can use Java classes, so the function body will just be
return java.lang.Math.sqrt(x);