Retrieval-Augmented Generation Must Move Beyond Factual Grounding to Represent Diverse Opinions
Standard Retrieval-Augmented Generation (RAG) systems are biased toward factual accuracy, ignoring subjective opinions. The paper proposes Opinion-Aware RAG (O-RAG), a new architecture that enriches documents with opinion attributes before indexing. This shift moves the goal from minimizing uncertainty to preserving genuine heterogeneity in retrieved content, aiming to prevent echo chambers and ensure fair representation.