Retrieval‑augmented generation (RAG) systems rely on external corpora, yet these sources may contain outdated, contradictory, noisy, or unreliable documents, which harms answer reliability. Prior work has used document relations to improve reliability, but propagation is limited to directly compared document pairs. To overcome this, we introduce TrustPropRAG. The approach models document relations as a graph and estimates a trust score for each document via multi‑hop propagation across the graph.
TrustPropRAG anchors the propagation with a small set of human feedback on document reliability, extending these costly signals to the whole corpus. Concretely, given the constructed document relation graph, we formulate an optimization problem that jointly captures pairwise document relations and user feedback; solving it yields a trust score per document. These scores guide the selection of reliable documents and enable trust‑aware answer generation.
Evaluation shows that TrustPropRAG improves both retrieval quality and exact‑match accuracy over baselines, and remains robust when feedback is sparse or noisy.
Review