EUVIMEDEUVIMED entdecken
Crossref

Replacing Multi-Step Assembly of Data Preparation Pipelines with One-Step LLM Pipeline Generation for Table QA

Fengyu Li, Junhao Zhu, Kaishi Song, Lu Chen, Zhongming Yao, Tianyi Li, Christian S. Jensen

Proceedings of the VLDB Endowment · 01.05.2026 · 19 (9) · 2372-2384

DOI: 10.14778/3819518.3819557 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the VLDB Endowment, online erschienen am 19.08.2026. Verfasst von Fengyu Li und 6 weiteren Autor:innen. Die Arbeit nennt 44 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Table Question Answering (TQA) aims to answer natural language questions over structured tables. Large Language Models (LLMs) enable promising solutions to this problem, with operator-centric solutions that generate table manipulation pipelines in a multi-step manner offering state-of-the-art performance. However, these solutions rely on multiple LLM calls, resulting in prohibitive latencies and computational costs. We propose Operation-R1, the first framework that trains lightweight LLMs (e.g., Qwen-4B/1.7B) via a novel variant of reinforcement learning with verifiable rewards to produce high-quality data-preparation pipelines for TQA in a single inference step. To train such an LLM, we first introduce a self-supervised rewarding mechanism to automatically obtain fine-grained pipeline-wise supervision signals for LLM training. We also propose variance-aware group resampling to mitigate training instability. To further enhance robustness of pipeline generation, we develop two complementary mechanisms: operation merge, which filters spurious operations through multi-candidate consensus, and adaptive rollback, which offers runtime protection against information loss in data transformation. Experiments on two benchmark datasets show that, with the same LLM backbone, Operation-R1 achieves average absolute accuracy gains of 8.83 and 4.44 percentage points over multi-step preparation baselines, with 79% table compression and a 2.2× reduction in monetary cost.

Zitieren

APA

Li, F., Zhu, J., Song, K., Chen, L., Yao, Z., Li, T., & Jensen, C. S. (2026). Replacing Multi-Step Assembly of Data Preparation Pipelines with One-Step LLM Pipeline Generation for Table QA. Proceedings of the VLDB Endowment, 19(9), 2372-2384. https://doi.org/10.14778/3819518.3819557

Vancouver

Li F, Zhu J, Song K, Chen L, Yao Z, Li T, et al. Replacing Multi-Step Assembly of Data Preparation Pipelines with One-Step LLM Pipeline Generation for Table QA. Proceedings of the VLDB Endowment. 2026;19(9):2372-2384. doi:10.14778/3819518.3819557

BibTeX
@article{Li2026,
  author = {Li, Fengyu and Zhu, Junhao and Song, Kaishi and Chen, Lu and Yao, Zhongming and Li, Tianyi and Jensen, Christian S.},
  title = {Replacing Multi-Step Assembly of Data Preparation Pipelines with One-Step LLM Pipeline Generation for Table QA},
  journal = {Proceedings of the VLDB Endowment},
  year = {2026},
  volume = {19},
  number = {9},
  pages = {2372--2384},
  doi = {10.14778/3819518.3819557}
}
Kostenlos starten →