The community member has a question about using an existing Knowledge Graph for RAG (Retrieval Augmented Generation). They expected each graph node to have multiple properties, such as name and description, which could be used as chunks of information during retrieval. However, it seems that only the node name and relationships are being retrieved, making the nodes "lightweight".
In the comments, another community member responds that as of now, they don't think it's possible to specify multiple properties to be returned from each node, unless the community member is doing something like "text-to-cypher". The commenter mentions that the knowledge graph support is "pretty basic" and open to improvements, but they haven't found it useful enough for production use-cases, especially when a knowledge graph is not already built.
The original community member thanks the commenter for the information and says they will look at the code to see if they can make a contribution.
I've got a question re: using an existing Knowledge Graph for RAG: The way I imagined it to work is that each graph node would have several properties (eg name, description, etc) with properties such as description playing the role similar to a chunk in a vectorDB, and retrievals would retrieve all properties. However, the way it seems to work is that only retrieves the name and relationships, ie the nodes are really lightweight. Is there a way to specify multiple properties to be returned as chunks of information from each node?
Hmm, as of right now, I don't think so, unless you are doing something like text-to-cypher
Our knowledge graph support is pretty basic tbh. Very open to any improvements. In general, I still haven't found it useful enough for production use-cases, especially when you don't already have a knowledge graph built
