What's the rational behind ReLU(s + o) − (s − o)?


> Practically, the entity-predicate matching is shown below:
$F(s, o) ≈ p$
where $F(s, o)$ is defined as: $ReLU(s + o) − (s − o)^2$

I don't see the reason why the equation must be this way. If (s-o) refers to the difference between the 2 vectors, (s+o) refers to the combination of the 2 vectors, what does the equation $F(s, o)$ means?