AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Tools are under-documented: Simple Document Expansion Boosts Tool Retrieval

Xuan Lu, Haohang Huang, Rui Meng, Yaohui Jin, Wenjun Zeng, Xiaoyu Shen

ICLR 2026 · Large language models › Retrieval and knowledge · approach: Retrieval-augmented reasoning

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
train retriever on expanded tool documents instead of original (ablation)Qwen3-Embedding-4B · Tool-DENDCG@10: +3.67 -> +6.69 over Qwen3-Embedding-0.6B baseline (4B tool-embed)Improved
train reranker on expanded documents instead of original (ablation)Qwen3-Reranker-4B · Tool-DE / ToolRetNDCG@10: +2.00 -> +4.21Improved
evaluate BM25s on expanded documentsBM25s · Tool-DE vs ToolRetNDCG@10: 36.41 -> 39.35Improved
evaluate e5-mistral-7b-instruct on expanded documentse5-mistral-7b-instruct · Tool-DE vs ToolRetNDCG@10: 40.02 -> 38.85Hurt
evaluate Qwen3-Embedding-4B/8B on expanded documentsQwen3-Embedding-4B/8B · Tool-DE vs ToolRetRecall@10: R@10 -1.23 and -0.69, N@10 slightly upMixed
remove example_usage field from expanded documents (ablation)GritLM / BM25s · Tool-DENDCG@10Improved
remove function_description or tags from expanded documents (ablation)GritLM / BM25s · Tool-DENDCG@10Hurt
add example_usage field alone to original document (ablation)GritLM · Tool-DENDCG@10Hurt
add reranker on original (non-expanded) ToolRet documentsjina-reranker-m0 · ToolRetNDCG@10: -0.08No clear effect
add reranker on expanded documentsQwen3-Reranker-4B · Tool-DE vs ToolRetNDCG@10: +0.50 -> +1.20Improved
document expansion at evaluation time lowers similarity scores asymmetricallyGritLM · Tool-DEmean similarity: negatives Web -0.0055 vs positives -0.0014Improved
apply document expansion to tool documentsLLM-audited documents · 300 sampled docsincomplete-doc rate: 41.6% -> 23.5%Improved

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