Knowledge graph
To construct one automatically, one would need to do Named entity recognition, Entity linking, and then Relation extraction.
Integration of LLMs with knowledge graphs
To combine LLMs with knowledge graphs, we need to be able to, given the context and question, identify the relevant knowledge from the graph and then use those knowledge elements with the question and context to jointly reason.
Yasunaga2021QA-GNN proposes QA GNN method, which involves relevance scoring of knowledge nodes by LLM, creating a joint graph, and use Graph neural network to update their representations.