[PHOTON1233]

ITs very much a Knowledge graph that builds upon itself as the affinities (node and relationship triples) pile up from user interactions, like for example. User: Sam and I are thinking of making our relationship official.... Affinities picked up: Girlfriend---OFFFICIALY_DATING--->Sam.

Keywords are soo 2022, semantic similarity is still employed to find pre-existing nodes and their relationships (affinities), and if not, new affinities are made, expanding the user's personal KG.

Its fun to build and maybe something can come out of it. Thanks for the comment!

[compressedgas]

I thought so. It is just that "keyword search on graphs" is the name of the search problem even if something other than keyword matching is used to pick which nodes in the graph to find a minimal connecting graph and a central result node from.

[PHOTON1233]

Thats pretty much it. There's also building on top of an existing graph and making sure new nodes are coherent and well connected to surrounding nodes which is a much tougher can of worms. Also, Navigating a KG (without multiple llm-hoping) is another challenge that needs to be solved for 'working-out retrieval'. Something that can navigate a forest of dividing paths without resorting to an llm call. Thanks for the name "keyword search on graphs". I found some great articles on the topic.