Authors: Xiaokang Zhang, Jing Zhang, Zeyao Ma, Yang Li, Bohan Zhang, Guanlin Li, Zijun Yao, Kangli Xu, Jinchang Zhou, Daniel Zhang-Li, Jifan Yu, Shu Zhao, Juanzi Li, Jie Tang

Published on: March 28, 2024

Impact Score: 7.8

Arxiv code: Arxiv:2403.19318

Summary

• What is new: TableLLM introduces a novel distant supervision method for training that focuses on tabular data, equipped with a reasoning process extension strategy and a cross-way validation strategy.
• Why this is important: Existing LLMs struggles with handling tabular data in real-world office documents and spreadsheets efficiently.
• What the research proposes: Developing TableLLM, a 13 billion parameter LLM, designed specifically for tabular data manipulation tasks and trained using a new distant supervision technique.
• Results: TableLLM outperforms both general-purpose and tabular data-focused LLMs in handling document and spreadsheet format tasks, verified via a new benchmark and evaluation pipeline.

Technical Details

Technological frameworks used: nan

Models used: Large Language Models (LLMs), TableLLM with 13 billion parameters

Data used: Automatically generated data validated through cross-way validation strategy

Potential Impact

Business productivity software, Data analysis tools, Spreadsheet and document handling applications

Want to implement this idea in a business?

We have generated a startup concept here: TableMastery.