Source Attribution in Retrieval-Augmented Generation
It addresses the challenge of providing explainability for RAG systems, which is crucial for users relying on LLM outputs, though it is incremental by applying existing attribution techniques to a specific domain.
This paper tackles the problem of attributing the influence of retrieved documents in Retrieval-Augmented Generation (RAG) systems by adapting Shapley-based methods, comparing them with approximations to reduce computational costs from expensive LLM calls, and evaluating their effectiveness in identifying critical documents under complex relationships.
While attribution methods, such as Shapley values, are widely used to explain the importance of features or training data in traditional machine learning, their application to Large Language Models (LLMs), particularly within Retrieval-Augmented Generation (RAG) systems, is nascent and challenging. The primary obstacle is the substantial computational cost, where each utility function evaluation involves an expensive LLM call, resulting in direct monetary and time expenses. This paper investigates the feasibility and effectiveness of adapting Shapley-based attribution to identify influential retrieved documents in RAG. We compare Shapley with more computationally tractable approximations and some existing attribution methods for LLM. Our work aims to: (1) systematically apply established attribution principles to the RAG document-level setting; (2) quantify how well SHAP approximations can mirror exact attributions while minimizing costly LLM interactions; and (3) evaluate their practical explainability in identifying critical documents, especially under complex inter-document relationships such as redundancy, complementarity, and synergy. This study seeks to bridge the gap between powerful attribution techniques and the practical constraints of LLM-based RAG systems, offering insights into achieving reliable and affordable RAG explainability.