Yes, and what I find even more worrying about a densely connected graph is that it is not easy to extract information! Lack of connections (factorizing knowledge) is as important as noticing connections. Otherwise you'll be swimming in connections of marginal relevance (i.e. distractions). IMHO a sprinkling of connections, resulting in a relatively sparse graph is quite easy to make sense of. It's not easy to foresee where the optimum might be.
You're absolutely correct, that is the dominant hyperparameter for a tool like this. If every word was hyperlinked, it would be nonsense. The good news is we don't need to analyze and forecast what the optimum is, we just need to ride the bike and learn to balance.
