I would recommend the first approach, to use edge predicates. The main advantage that this gives you is being able to find all nodes that have a particular label (by following the edges out of the label node). Other than this aspect, the performance of both solutions should be about the same.
1 Like