Question about rerankers in knowledge graphs

I was just thinking about trying out a new RAG approach that works similarly to LightRAG, but I can’t find anything about it online.

The documents or text sections are stored as nodes in a knowledge graph. A reranker is then used to retrieve documents based on a query. Subsequently, the query itself becomes a new node that immediately establishes a connection with the most similar nodes identified by the reranker.

This might offer the advantage that rerankers are more precise than, for instance, contextual RAG. In contrast, LLMs used for node extraction can only generate specific labels or keywords for nodes, which might be too coarse-grained.