Problem Statment

I’m building a feature to create a diagram that maps the relationships among a lot of natural language sentences.

I think this requires the extraction of relationships among a lot of natural language statements. “Relationship” in this context is loosely defined. It is a sort of relevance, for example, the relationship between (a) “Work hard on things you love, in fact, excel it.”, (b) “Try to work hard with people you like to avoid burning out.” and (c) “Hard work compounds like interest.” could be that (a) is supported by (b) and (c). So in a diagram, (b) and (c) would be linked to (a). Relationships could also be “agreement”, “cause and effect”, “premise and conclusion”.

Yet, this is not the construction of a mind map. There aren’t necessary hierarchies among ideas. Also, how the statements link together depends on how you view it, so you can link it however you want as long as it makes sense.

What I’ve tied

I’ve tried creating cards and presenting a screenshot of all of them to LLM visual models to identify them, and then outputting an adjacent matrix containing edges, nodes and relationship attributes. But the result is not satisfying.

Are there any other solutions?