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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| 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 -> 4722 | Improved |
| 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 -> 4722 | Improved |
| add execution feedback to clause-wise non-nested generation (ablation) | Clause generation · Non-nested query generation (NBA) | total LLM calls: 1316 -> 1134 | Improved |
| use execution-guided instead of one-shot for non-nested queries (ablation) | One-Shot generation · Non-nested query generation (NBA) | duplicate queries: 1265 -> 27 | Improved |
| provenance repair with smaller LLM (ablation) | gpt-oss-20B · Nested generation cost | total LLM calls: 5977 -> 4854 | Improved |
| provenance repair with larger LLM (ablation) | gpt-oss-120B · Nested generation cost | total LLM calls: 8425 -> 3831 | Improved |
| use GPT-5 instead of GPT-3.5-turbo backbone | ODYSSEY · SPARTA (Oracle) average | F1: 20.6 -> 35.6 | Improved |
| use GPT-5 instead of GPT-3.5-turbo backbone | HProPro · SPARTA (Oracle) average | F1: 20.3 -> 40.4 | Improved |
| evaluate on SPARTA instead of HybridQA | ODYSSEY with GPT-5 · Oracle table-text QA | F1: 69.5 -> 35.6 | Hurt |
| evaluate on SPARTA instead of HybridQA | HProPro with GPT-5 · Oracle table-text QA | F1: 70.5 -> 40.4 | Hurt |
| require retrieval instead of oracle evidence | HELIOS + HProPro with GPT-5 · SPARTA | F1: 40.4 -> 22.6 | Hurt |
| require unstructured-text cross reasoning instead of table-only | HProPro with GPT-5 · SPARTA (Oracle) | F1: 45.2 -> 16.3 | Hurt |
Closest papers by meaning
- CofCA: A STEP-WISE Counterfactual Multi-hop QA benchmark (ICLR 2025 · similarity 0.89)
- TableBench: A Comprehensive and Complex Benchmark for Table Question Answering (AAAI 2025 · similarity 0.89)
- Text2Analysis: A Benchmark of Table Question Answering with Advanced Data Analysis and Unclear Queries (AAAI 2024 · similarity 0.88)
- TrustUQA: A Trustful Framework for Unified Structured Data Question Answering (AAAI 2025 · similarity 0.88)
- Diagnosing and Addressing Pitfalls in KG-RAG Datasets: Toward More Reliable Benchmarking (NeurIPS 2025 · similarity 0.88)
- Tree-of-Reasoning Question Decomposition for Complex Question Answering with Large Language Models (AAAI 2024 · similarity 0.87)
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.
