AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Hierarchical Abstract Tree for Cross-Document Retrieval Augmented Generation

Ziwen Zhao, Menglin Yang

ICML 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Layout-aware document understanding · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace k-means-type tree (RAPTOR) with Psi-RAG abstract tree and agentic hybrid retrievalTree-RAG · Single/multi-hop QA averageF1 (%): 36.69 -> 62.77Improved
Psi-RAG versus HippoRAG 2 on cross-document QAPsi-RAG · Single/multi-hop QA averageF1 (%): 55.41 -> 62.77Improved
add retrieval and answering agent (ablation)Psi-RAG · multi-hop QAF1 (%): average gain of ~20%Improved
add R&A agent (ablation)Psi-RAG · NQF1 (%): +1.89%Improved
add sparse keyword retrieval (BM25) (ablation)Psi-RAG · PopQAF1 (%): +21.82%Improved
add sparse keyword retrieval (BM25) (ablation)Psi-RAG · 2WikiF1 (%): +41.21%Improved
use Qwen3-Reranker-8B instead of RRF for fusion (ablation)Psi-RAG · QA datasetsF1 (%)No clear effect
add query reorganization (ablation)Psi-RAG · MuSiQueF1 (%): 45.82 -> 47.83Improved
add query reorganization (ablation)Psi-RAG · MultiHop-RAGF1 (%): 55.03 -> 55.18No clear effect
use summative instead of keyword abstracts (ablation)Psi-RAG · Single/multi-hop QA averageF1 (%): 62.38 -> 62.77No clear effect
replace Llama-3.3-70B abstraction agent with Llama-3.2-1B (ablation)Psi-RAG · HotpotQAF1 (%): 74.85 -> 73.23No clear effect
replace Llama-3.3-70B R&A agent with Gemma-3-27B (ablation)Psi-RAG · 2Wiki and MuSiQueF1 (%): 2Wiki 76.94 -> 71.78; MuSiQue 47.83 -> 49.30Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy