I've been thinking about this all morning and have come up with this:

When the link is the same, we can just union on the thread and sort it as any other thread.

When the link is not the same, but the content is similar, I wonder if you could run a semantic comparison (I'm only just starting to learn about this), and when the comparison scores above a certain threshold, union the threads.

Self-text posts might just make sense to leave alone.

I wonder how feasible it would be to set your instance up to aggregate content from duplicate communities.