AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

SPARTA: Scalable and Principled Benchmark of Tree-Structured Multi-hop QA over Text and Tables

Sungho Park, Jueun Kim, Wook-Shin Han

ICLR 2026 · Large language models › Retrieval and knowledge · approach: Explanatory and multiple-choice QA benchmarks · 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
add provenance-based repair to post-order nested query generation (ablation)Post-Order generation · Nested query generation (NBA, Llama-3.1-70B)total LLM calls: 8253 -> 4722Improved
use post-order with provenance instead of one-shot-k (ablation)nested query generation · Nested query generation (NBA, Llama-3.1-70B)total LLM calls: 13962 -> 4722Improved
add execution feedback to clause-wise non-nested generation (ablation)Clause generation · Non-nested query generation (NBA)total LLM calls: 1316 -> 1134Improved
use execution-guided instead of one-shot for non-nested queries (ablation)One-Shot generation · Non-nested query generation (NBA)duplicate queries: 1265 -> 27Improved
provenance repair with smaller LLM (ablation)gpt-oss-20B · Nested generation costtotal LLM calls: 5977 -> 4854Improved
provenance repair with larger LLM (ablation)gpt-oss-120B · Nested generation costtotal LLM calls: 8425 -> 3831Improved
use GPT-5 instead of GPT-3.5-turbo backboneODYSSEY · SPARTA (Oracle) averageF1: 20.6 -> 35.6Improved
use GPT-5 instead of GPT-3.5-turbo backboneHProPro · SPARTA (Oracle) averageF1: 20.3 -> 40.4Improved
evaluate on SPARTA instead of HybridQAODYSSEY with GPT-5 · Oracle table-text QAF1: 69.5 -> 35.6Hurt
evaluate on SPARTA instead of HybridQAHProPro with GPT-5 · Oracle table-text QAF1: 70.5 -> 40.4Hurt
require retrieval instead of oracle evidenceHELIOS + HProPro with GPT-5 · SPARTAF1: 40.4 -> 22.6Hurt
require unstructured-text cross reasoning instead of table-onlyHProPro with GPT-5 · SPARTA (Oracle)F1: 45.2 -> 16.3Hurt

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