Hierarchical Abstract Tree for Cross-Document Retrieval-Augmented Generation

Ziwen Zhao, Menglin Yang · 2026

Read Hierarchical Abstract Tree for Cross-Document Retrieval-Augmented Generation by Ziwen Zhao, Menglin Yang (2026) 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-2605-00529.txt.

Description

Ziwen Zhao and Menglin Yang improve retrieval-augmented generation for questions whose answers are scattered across many documents. Tree-based systems index documents hierarchically for questions at several levels of detail, but single-document designs scale badly: rigid clustering injects noise, separate trees never link documents, and coarse summaries bury fine detail. Their framework tackles all three faults.

Its index is a hierarchical abstract tree built by repeatedly merging and collapsing nodes, adapting to the data's own distribution instead of assuming one. A retrieval agent then works the tree with reorganized queries and a hybrid search, serving tasks from pinpoint answers to whole-document summaries. On cross-document multi-hop benchmarks, the system beat two respected rivals in average F1 score by clear margins. The authors release their code for reuse.

Catalog record

CREATORZiwen Zhao, Menglin Yang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE101 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.00529

Browse the full library · About the Tower of Babel