unstrucured2Graph: There is no url/filename in the dataset to explore the page

[Josipmrden]

I would think that url is good because LLM's can scrape. But we don't provide it and hence we restrict the LLM of doing that action.

Idea with Memgraph's documentation is either:

• either scrape the cypher queries in the documentation using an LLM
• index it by itself with additional ingestion mechanism, prompt

[antejavor]

I am not sure what this is? Is this solved with the docs MCP?

Can you add a bit context?

[gitbuda]

An additional idea here is to introduce something like "content type", for each URL what is that content about (docs pages, legal text, financial statement, essey, blog, research paper), in other words, what's the top-level semantic. That would allow specific parsing pipelies, beucause it's possible to assume some elements and structure.