Performance of a simple SPARQL query using filters

Question

I have a simple SPARQL query that looks for shared nodes within a graph using Filter, and Union arguments. The query take an unusually long time to compile. I was hoping to see if it is possible to rearrange this structurally to increase its performance.

The shared nodes are any of the three kinds of

1. shared between two objects
2. shared between two subjects, and
3. object of one triple while subject of another.

I provided the graph of the possible data in another question I asked.

The query looks as below:

    """SELECT DISTINCT ?b
       WHERE{ 
       {
          ?b ?p1 ?a.
          ?b ?p2 ?c.
          filter(?a != ?c).
      }
       UNION
       {
          ?a ?p1 ?b.
          ?c ?p2 ?b.
          filter(?a != ?c).
      }
       UNION
       {
          ?a ?p1 ?b.
          ?b ?p2 ?c.
          filter(?a != ?c).
       }}
       """

From since I posted this, I did some experiments and learned the most time consuming part is the middle one.

       {
          ?a ?p1 ?b.
          ?c ?p2 ?b.
          filter(?a != ?c).
      }