Yes we've examined it. I think our approach is focused on granular, opinionated schemas as the way to represent the information that you care about within your text, as opposed to trying to represent all the entities & relationships possible, including the ones you may not care about. This is as it is difficult to extract everything completely, and a dense network of relationships make it hard to accurately retrieve from that graph.

GraphRAG's 'communities' is basically clustering, which we think does not overlap well with KG RAG systems - I believe other folks agree with us: https://www.linkedin.com/posts/dselman_knowledge-graphs-rag-is-not-all-you-need-activity-7222236029276024832-5KUb?utm_source=share&utm_medium=member_desktop