Prompting Large Language Models for Table Question Answering


In this study, we propose methods that leverage large pretrained language models for table question answering. Results have shown that the methods are effective compared with finetuned models in this area.

Published on December 01, 2022 by Yihan

NLP

1 min READ

This post will talk about our recent research in the area of AI for data science. Data science and analytics are essential for various fields including business development, market research, investments and so on, The number of data analyst positions has also been increasing in recent years. Nevertheless, sometimes a data analytic problem could be very complicated for a SQL/Python novice. And, in the same time, some data analysts also want to save efforts in solving straight-forward problems. Noticing these problems, we gathered some recent works that might help data scientists in developing more efficient code. We also propose our own method, which is still under review. After publication, we will be able to talk more about our solution.


Introduction & Motivation

Current Methods

Training-free Methods

Prompting Codex/ChatGPT

Binder: Binding language models for table qa

End-to-end Methods

RL Methods