EnSI-RAG: Entity-Structure-Indexed Retrieval-Augmented Generation for Long-Document Question Answering
The paper proposes indexing long documents by entities and their structured attributes instead of raw text chunks.
EnSI-RAG stores records linking an entity, its type, a semantic category, and a value back to source passages. At query time, those records guide retrieval, while an LLM uses the retrieved passages to produce the answer. The authors report 78.24 average accuracy across Loong and Oolong, 6.62 points above their referenced published baselines. Code is listed as available. ArXiv · AI/CL/LG's note
EnSI-RAG stores records linking an entity, its type, a semantic category, and a value back to source passages. At query time, those records guide retrieval, while an LLM uses the retrieved passages to produce the answer. The authors report 78.24 average accuracy across Loong and Oolong, 6.62 points above their referenced published baselines. Code is listed as available. ArXiv · AI/CL/LG's note
score 4