🤖 AI Summary
This work addresses the limitations of existing retrieval-augmented generation methods, which struggle to simultaneously support structured constraints and multi-hop reasoning, as well as knowledge graph–based approaches that suffer from semantic fragmentation, high maintenance costs, and difficulty in updating. The authors propose the SAG architecture, which organizes documents via an event–entity index, preserving n-ary relations within original text blocks without constructing a global knowledge graph. At query time, SAG dynamically connects relevant event blocks through shared entities to form an evidence neighborhood. A novel dynamic hyperedge mechanism is introduced to avoid decomposing relations into triples, thereby maintaining semantic integrity while enabling efficient incremental updates and complex reasoning. By integrating SQL-style structured retrieval with dynamic connection, SAG achieves state-of-the-art performance on HotpotQA, 2WikiMultiHopQA, and MuSiQue, attaining a Recall@5 of 80.36% on MuSiQue—11.52 percentage points above the strongest baseline.
📝 Abstract
While retrieval-augmented generation (RAG) has proven effective at giving LLMs access to external knowledge, mainstream dense-retrieval implementations remain inherently limited in handling structured constraints and multi-hop reasoning. Graph-based methods address this by constructing knowledge graphs offline, but they often fragment semantics, incur high maintenance, and complicate incremental updates. We propose SAG (SQL-Retrieval Augmented Generation), a structured retrieval architecture that organizes documents into an event-entity index without building a global knowledge graph. SAG represents each chunk as a semantically complete event paired with its entities, forming a latent hyperedge that preserves n-ary relations without decomposing them into triples. At query time, SAG treats shared entities as join keys to connect related chunks. This dynamically yields a query-scoped neighborhood of events, and yet every piece of evidence remains the original chunk throughout. Experiments on HotpotQA, 2WikiMultiHopQA, and MuSiQue show that SAG achieves the best retrieval and end-to-end QA performance on every benchmark, with gains that widen as reasoning-chain complexity increases. On MuSiQue, where multi-hop evidence chaining is most demanding, SAG reaches 80.36% Recall@5, outperforming the strongest baseline by 11.52 points. This work paves the way for knowledge infrastructure that enables LLM agents to retrieve and reason over continually growing organizational knowledge.