SAGE : A Framework of Precise Retrieval for RAG Guoliang Li is the corresponding author. This paper was supported by National Key R&D Program of China (2023YFB4503600), NSF of China (61925205, 62232009, 62102215), Huawei, TAL education, and Beijing National Research Center for Information Science an
Jintao Zhang, Guoliang Li *, Jinyang Su · 2025
Read SAGE : A Framework of Precise Retrieval for RAG Guoliang Li is the corresponding author. This paper was supported by National Key R&D Program of China (2023YFB4503600), NSF of China (61925205, 62232009, 62102215), Huawei, TAL education, and Beijing National Research Center for Information Science an by Jintao Zhang, Guoliang Li *, Jinyang Su (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-01713.txt.
Description
Jintao Zhang, Guoliang Li, and Jinyang Su present a sharper search frame for chat models that read from a fixed corpus, in a paper from 2025. Such models fail at questions most often because search pulls the wrong chunks, not because the model reasons poorly. Two flaws drive the misses. Cuts split the corpus with no heed to sense, and fixed chunk counts trade missed facts for noise.
The frame first trains a cutter that splits text at sense bounds. A picker then adds chunks while their fit scores fall fast and stops when they fade. Last, the model itself judges whether the haul is too fat or too thin and trims or tops it up. Tests lift answer quality by a wide margin and cut token cost near in half.
Catalog record
| CREATOR | Jintao Zhang, Guoliang Li *, Jinyang Su |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-01713