In this case at least, they're going from a peak of ~1.66 posts/day to a valley of ~1.54. As an aggregate that looks like a compelling pattern, but consider it on an individual level: we're talking about one post every ten days. Initially looking at the graph made me think the difference was way greater than a post every ten days, so I'd agree with the GP that it is misleading; I was mislead.
Looking at the graph, it seems like we're talking about ~16.5 posts in the ten days before a relationship versions versus ~15.7 in the ten days after. In the context of an individual, Facebook might be able to combine this with a bunch of other data to make guesses about people entering relationships, but a quick look at the graph makes it look like they could do it from just this one piece of data.
FWIW, I think people are often too quick to dismiss graphs as misleading due to the y-scale and I do think the graph is interesting. That said, I think a lot depends on the context, and being directly below "When You [emphasis added] Fall in Love, This Is What Facebook Sees" makes this seem misleading.