Optimizing Compound Retrieval Systems

Harrie Oosterhuis, Rolf Jagerman, Zhen Qin, Xuanhui Wang · 2025

Read Optimizing Compound Retrieval Systems by Harrie Oosterhuis, Rolf Jagerman, Zhen Qin, Xuanhui Wang (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-2504-12063.txt.

Description

Harrie Oosterhuis, Rolf Jagerman, Zhen Qin, and Xuanhui Wang widen in 2025 the idea of a compound search system, where many rankers work as one. Most search stacks run models in a cascade, each re ranking a shrinking short list. Other teamwork forms exist, and large language models can now judge pairs of results head to head.

Their frame learns where to apply each model and how to blend its calls into one ranking. A classic word ranker pairs with a language model's pair judgments in their tests. The blend is tuned against a set quality and cost target. Optimized designs beat plain cascades in their trials.

Catalog record

CREATORHarrie Oosterhuis, Rolf Jagerman, Zhen Qin, Xuanhui Wang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE58 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-12063

Browse the full library · About the Tower of Babel