Attribution Techniques for Mitigating Hallucination in RAG-based Question-Answering Systems: A Survey

This is the authors' abstract. We don't add key points for this paper.

Large Language Models (LLMs)-based question answering (QA) systems play a critical role in modern AI, demonstrating strong performance across various tasks. However, LLMgenerated responses often suffer from hallucinations-unfaithful statements lacking reliable references. Retrieval-Augmented Generation (RAG) systems enhance LLMs by retrieving external references to inform responses, but they can also introduce new forms of hallucination due to complex interactions between the retriever and generator components. Research has increasingly focused on attributionbased techniques within RAG systems to mitigate these hallucinations. However, a unified pipeline for implementing these techniques, along with a clear taxonomy and systematic comparison of their strengths and weaknesses, is still lacking. This survey aims to address this gap by: (i) outlining a taxonomy of hallucination types in RAG systems, (ii) presenting a unified pipeline for attribution techniques, (iii) reviewing these techniques based on the hallucinations they target, and (iv) discussing their strengths and weaknesses with practical application guidelines. This work provides valuable insights for future research and practical use of attribution techniques in RAG systems.