Very cool! I think only using nouns and named entities as nodes is somewhat limiting and in information extraction we can do better, using concept embeddings w/ similarity links for instance. But that might just be an improvement for improvement sake. This looks like solid software engineering groundwork. I will try this on my notes, writing and literature collection and I might put in some fancier NLP.