Self-Retrieval: End-to-End Information Retrieval with One Large Language Model
This addresses the problem of separated architectures in IR systems for researchers and practitioners, offering a novel integration approach that is not incremental but represents a new paradigm.
The paper tackles the limited integration of large language models (LLMs) in information retrieval (IR) systems by introducing Self-Retrieval, an end-to-end LLM-driven architecture that unifies IR functions within a single LLM, resulting in significant performance improvements over existing retrieval approaches and enhanced downstream applications like retrieval-augmented generation.
The rise of large language models (LLMs) has significantly transformed both the construction and application of information retrieval (IR) systems. However, current interactions between IR systems and LLMs remain limited, with LLMs merely serving as part of components within IR systems, and IR systems being constructed independently of LLMs. This separated architecture restricts knowledge sharing and deep collaboration between them. In this paper, we introduce Self-Retrieval, a novel end-to-end LLM-driven information retrieval architecture. Self-Retrieval unifies all essential IR functions within a single LLM, leveraging the inherent capabilities of LLMs throughout the IR process. Specifically, Self-Retrieval internalizes the retrieval corpus through self-supervised learning, transforms the retrieval process into sequential passage generation, and performs relevance assessment for reranking. Experimental results demonstrate that Self-Retrieval not only outperforms existing retrieval approaches by a significant margin, but also substantially enhances the performance of LLM-driven downstream applications like retrieval-augmented generation.